January/February 2003

Seismologists Must Begin Forecasting Earthquakes

For about the past year, I have been engaged in a behavior that some seismologists would consider ridiculous, others foolish, others amusing, and still others a waste of time. What could I have been doing that would raise such opinions in so many colleagues? Well, I have been issuing regular and frequent earthquake forecasts. I have been doing this publicly for New England and privately to a few colleagues for California. So far, there has yet to be a time when an earthquake that I have forecast has actually occurred (well, maybe one). However, that has not discouraged me. I intend to keep trying.

So what would cause an otherwise conservative and mainstream seismologist to engage in such rash and foolish behavior? I do it because I think that earthquake forecasting and prediction must become a major direction of research and public education for earthquake seismologists. The public expects us to predict earthquakes, as is known by anyone who is frequently interviewed about earthquakes. The work of many different branches of our government would be enhanced by a proven ability to predict damaging earthquakes. The earthquake research that stands to affect the financial markets is arguably earthquake prediction, since knowledge of earthquake risk is important for financial planning. As Chip Groat, Director of the U.S. Geological Survey, indicated at the 2002 annual SSA meeting, earthquake forecasting and prediction must be important goals if we are to improve future funding for seismological science in the U.S.

I believe that the seismological community needs to be prominently promoting earthquake forecasting and prediction research. In particular, we need to convince the public that much of our current research is helping us forecast future earthquake activity. I think some new earthquake forecasting and prediction research is needed, but more importantly I think we need to become more proactive in disseminating to the general public our best assessments of the spatial and temporal probabilities of future earthquake activity. I must say here that I am not simply talking about issuing time-independent seismic hazard maps or other time-independent assessments of future earthquake activity. Rather, I am talking about using our ability to respond to new earthquake occurrences and geophysical measurements to issue regular and perhaps frequent estimates of the probabilities of future earthquake activity. Just as the public can tune into a TV station or read a newspaper to get the latest weather forecast, so should there be outlets where the public can get the latest earthquake forecasts.

Right now, I think too many seismologists have attitudinal problems both with how we regard our own research and how we present our research to the public. Going back at least to the days of Charles Richter more than a half-century ago, seismologists have usually regarded with suspicion those who claim to be predicting earthquakes. When a seismologist makes a prediction of a large earthquake, the natural scientific conservatism of the seismological community takes over in the evaluation of the prediction. The prediction is not immediately accepted, but rather the scientific process demands that the prediction must be carefully studied and evaluated before any pronouncements are made about it. This is not unlike the connoisseur of a fine wine, who must smell and taste the contents of a bottle before judging how fine the wine really is. After the heady earthquake prediction days of the 1970's, many seismologists today are suspicious of earthquake predictions by scientists. Too many proposed methods to predict earthquakes have failed to produce positive results, in many cases because there was an inadequate scientific basis for the prediction methods. I think it is now commonly recognized that accurate scientific predictions of large earthquakes are a notoriously difficult business, and we seismologists have tried to make this point very clear to the public at large.

But with the public thirst for earthquake forecasting and prediction, fueled in part by the increasingly accurate weather forecast capabilities that are displayed prominently in our media, society wants seismologists to make progress toward predicting earthquakes. Many totally baseless predictions of major earthquakes have been published in sensationalized supermarket tabloids and other potboiler literature. Unfortunately, too often the public has taken these predictions seriously. I believe this is because there is almost no counterpart routine forecasting or prediction effort from the scientific community. Things are clouded even further when pseudoscientific earthquake predictions are made, like the 1990 prediction by Iben Browning of an imminent large earthquake in the New Madrid seismic zone. The public is prone to take these as credible scientific earthquake forecasts, which they patently are not.

I believe the seismological community has significantly undersold its capabilities regarding earthquake time-dependent forecasts, at least for the small, routine earthquake activity that plays itself out year in and year out. I think this is partly true because even the experts tend to confuse earthquake forecasting and prediction. Earthquake forecasting involves making a statement concerning the possibility of a future earthquake where the forecast has wide uncertainty in any or all of the time, magnitude, and location of the coming event. On the other hand, an earthquake prediction is a statement concerning the possibility of a future earthquake where the statement has a very narrow uncertainty in the time, magnitude, and location of the coming event. I think things are further confused because we tend to think of earthquake forecasting and prediction only for the large, damaging earthquakes, as though it is not necessary to forecast or predict small events that are incapable of causing damage.

In fact, I think that seismologists already have the capabilities to make time-dependent earthquake forecasts of routine seismicity in some specific cases. For example, Omori's law gives us an empirical tool to estimate the rate of aftershocks with time after a mainshock, and a quantitative model for the prediction of aftershock rates with time and magnitude has already been derived by Lucy Jones and Paul Reasenberg. Static stress changes following earthquakes apparently increase the probabilities of aftershocks in some areas and decrease the probabilities in other areas, as has been shown by many investigators. Distant triggering of seismicity by large earthquakes is becoming recognized in more and more cases. Even routine regional seismic network monitoring has shown that areas can undergo periods of increased seismicity, as was the case in the Los Angeles area in 1989. In New England during the period from 1975 to 1999, Alan Kafka and I found that when a felt earthquake took place in New England, there was an enhanced probability that another felt earthquake would take place somewhere in New England over the next week or so. The location of the second earthquake did not appear to depend at all on the location of the first earthquake (i.e., it could be anywhere in New England). We published this research in the June 2002 issue of BSSA. Recent work by Kafka and myself (so far unpublished) has indicated that there may be a similar non-Poissonian temporal clustering of M 4+ earthquakes in eastern and central California.

So why do I believe that the public is not better informed about these research results? I feel that we seismologists do not do a very good job of presenting such results to the public at large. Just as weather forecasters give us daily (even hourly) updates on the upcoming weather, so I believe seismologists should be giving regular updates on the probabilities of future earthquakes. These updates should be done on a frequent basis-perhaps not daily but no less frequently than biweekly in a seismically active area like California. Clearly, this can only be done if the forecasts focus primarily on the smaller, routine seismicity, that which takes place with the greatest frequency. One can argue that these forecasts are uninteresting, but I disagree. Weather forecasters do not only tell us about severe storms; they also give forecasts when the weather is tranquil. If one is forecasting the probabilities of the numbers and spatial locations of future earthquakes of, say, M 2.5+ over a short period (such as a week or two), I think such forecasts would change frequently enough to be of steady interest to the public. Obviously, aftershock areas of larger earthquakes would show up on such forecasts. The same would happen for earthquake swarms. Regional increases or decreases in the background seismicity rate might also be reflected in the forecasts.

Weather forecasters are not afraid to give probability values in their forecasts. The general public can understand that a 30% chance of rain means that precipitation is less certain than a 70% chance of rain. I think forecasts of possible changes in routine earthquake activity can be given in terms of probabilities or changes in probabilities of coming earthquake activity. Since last year, Weston Observatory has been giving weekly estimates of the seven-day probability of a felt earthquake in New England on the Weston Observatory Web site (http://www.bc.edu/westonobservatory and follow the links to the earthquake probability Web page). Right now, the probability levels range from 11% if no felt earthquake took place in New England during the past seven days to 35% if an M 3.5+ event has just occurred in the region. These probability levels include the possibilities of aftershocks at the site of the initial event. The Weston Observatory Web pages also contain a map of areas most likely to be the epicenter of the next M 2.7+ event. This map is based on Alan Kafka's work in the November/December 2002 issue of SRL to quantify the spatial repeatability of earthquake epicenters. Everyone who works with seismicity knows that most routine earthquakes happen where earthquakes have happened in the past, and just a few years of seismic monitoring usually reveal the seismically active areas. Kafka's contribution has been to quantify this qualitative observation.

The probability levels for New England are easy to relate to statistics that many are familiar with. For example, the best professional baseball player gets a hit about 35% of the times he bats, a mediocre player about 22% of the time, and a poor player (maybe a pitcher) only 11% of the time. Nomar Garciaparra of the Boston Red Sox can fail to get a hit two out of three times and still be considered a success. The same should be true of our New England earthquake forecasts if the future earthquakes follow the same statistical patterns as the past activity.

I think there are many advantages to issuing regular forecasts of the probabilities of future earthquakes greater than some minimum (and small) magnitude. First, the small magnitudes of most of the events for which the forecasts are made means that over the period of a few years, we will accumulate enough forecast cases that we can test in a quantitative, statistical way how well the forecasts are working. This contrasts to forecasts and predictions of M 6+ earthquakes, for which one might need to wait years or even decades to evaluate a single forecast in a seismically active region (witness the Parkfield, California prediction). A second advantage comes from making forecasts on the usually small earthquakes over short periods. While the public is typically interested in any forecast of an earthquake, there is no panic or hysteria upon the forecast of an increased probability of what will probably be a small earthquake in the next few days. This is no forecast of the end of the world. It is not easily sensationalized. Yes, a damaging earthquake (say M 6 or even M 7) can be consistent with such a forecast, but the odds heavily favor the earthquake magnitude being near or at the lower magnitude limit of the forecast. Finally, the public has a chance to become accustomed to what we know and what we don't know about coming earthquakes through such regular forecasts. If an event occurs in a time window identified as having an enhanced earthquake probability, the public will say we were successful. If it doesn't, we will say that the nonoccurrence was consistent with the probability in the forecast. Such a system of regular and frequent earthquake forecasts will give the public a chance to get a better feeling for the rhythm of earthquake occurrences. Yes, we might be criticized if we are wrong. This often happens to the weather people, but it does not stop them from continuing to forecast the weather. With many forecasts, it is easier to learn from mistakes and improve the forecasting methods.

I will admit that most of the models describing temporal or spatial changes in seismicity rates that I described earlier are observational and empirical in nature with no direct physical models as their basis. The best and most useful models are ones derived from the physical processes that describe earthquake generation. Today, the best weather forecasting is based on predictions from detailed numerical models of the atmosphere and its expected evolution, models that often require extensive computing power to evaluate. Before such models were developed in the 1970's, most weather forecasting was based on the so-called analog method. Current weather conditions as well as weather conditions from the recent past were compared to older weather observations to find the closest past analog of the current weather. The subsequent weather patterns observed in the past analog were then used as the forecast for the current weather. It strikes me that we currently are in a similar state with earthquake forecasting-looking for past analogs of our current seismicity. I feel confident that as time progresses, not only will our analog-based earthquake forecasts improve, but so also will our ability to make model-based earthquake forecasts.

Clearly, I am calling for some new research on earthquake probabilities, although I would argue the basis for much of the research has already been laid. Even more so, I am calling for the development of new public information products and services. Who develops these, what they show, and how they are disseminated must all be worked out. The products must be appealing and informative, and they must clearly lay out their limitations. Obviously, there must be cooperation among many in the seismological community to make my proposed ideas effective. On the other hand, I must point out that a number of different private and public organizations issue weather forecasts on a regular basis. In fact, it is through the comparison of such forecasts that longer-term weather forecast models (models that generate forecasts for a couple of weeks to a season or longer) are tested. Thus, testing different earthquake forecast methods (perhaps by different groups) should be an exercise that seismologists embrace.

Let me close by saying that I believe that a major justification of both the Advanced National Seismic System (ANSS) and the EarthScope initiatives is improved earthquake forecasting and prediction. The new instrumentation, data communication systems, and data analysis structures for regional monitoring of routine earthquake activity to be developed and employed nationally under ANSS are necessary to improve earthquake forecasting capabilities in all parts of the country. Especially for empirically based earthquake forecasting schemes, the ability to quickly distribute and process weak-motion as well as strong-motion earthquake data is necessary if fast and accurate forecasts are to be made immediately following the occurrence of an earthquake. The Plate Boundary Observatory and the San Andreas Fault Observatory at Depth projects under EarthScope are designed to address directly the need for better information to constrain the physical models of earthquake generation. In addition, the data from the Plate Boundary Observatory as well as from USArray will provide new information that can help constrain earthquake forecasting models.

The public wants us to predict earthquakes. Maybe we can't do that today, but we can give the public understandable, time-dependent earthquakes forecasts on a regular basis. I firmly believe this would make the public more willing to support our science.

John E. Ebel
Weston Observatory of Boston College

To send a letter to the editor regarding this opinion or to write your own opinion, contact the SRL editor by e-mail.

Posted: 24 February 2003