FOR an article in Science magazine, its title was blunt:
“Earthquakes cannot be predicted.” The text of the piece, which appeared in
March last year (vol 275, p 161), was just as uncompromising. The authors, four
well-known scientists specialising in the physics of the Earth’s interior, were
calling for nothing less than the end of a line of scientific endeavour.
Earthquake prediction, they proclaimed, is doomed to failure. Anyone who funds
it should look for other ways to spend the money. Game over.
To some geoscientists, this was familiar stuff. For most of the past ten
years the lead author, Robert Geller, chief seismologist at Tokyo University,
had been ripping into earthquake prediction research at conferences and in
journal articles, and his standard targets were ready with standard rebuttals to
this latest attack. Still, it was clear that Geller and his coauthors had
touched a sore spot. For scientists who study natural disasters, earthquakes
have always been the ones that got away. Hurricanes, tornadoes, floods, even
volcanic eruptions can be forecast, monitored and tracked, their basic
mechanisms observed and modelled with some degree of confidence. But earthquakes
have remained cryptic and slippery—earthborn Godzillas that spring from
the depths, strike at will and then rumble out of existence, taking their
secrets with them.
Geller’s fury was mainly roused by the gargantuan Japanese prediction
programme, which he says has swallowed up billions of yen over the past few
decades and produced nothing in return. In spite of this, the Japanese research
continues, along with work in Greece, Russia and China. But back in Geller’s
native country, the US, earthquake prediction is the research that dare not
speak its name. It peaked sometime between the death of Elvis Presley and the
advent of Elvis Costello and has been declining ever since, a victim of its
early excesses. Yet the prediction scientists still have some tricks up their
sleeves—if they’re allowed to keep playing. Prediction might not be the
only game in town, they argue, but the horrific cost of earthquakes makes it a
gamble worth taking.
Advertisement
Long-range forecasts
The critics of prediction are not calling for an end to earthquake research
per se. After all, it’s their line of work too. Nor do they object to using
science to forecast the future. They are quite comfortable with long-term
estimates of earthquake probabilities, such as the 30-year seismic hazard maps
that the US Geological Survey prepares for 20 different segments of the San
Andreas Fault. Such efforts (which most experts now prefer to call earthquake
forecasts) are based on measurements of ground movement along faults, historical
records of earthquakes and evidence of ancient landslides, toppled trees and the
like—scientifically sound methods, as even the fiercest opponents of quake
prediction agree.
What draws their fire are efforts to make short-term earthquake predictions,
hours to months in advance. The idea is that earthquakes may be heralded by
warning signs—precursors powerful enough to be detected, sensitive enough
to distinguish a big earthquake from a small one and dependable enough to catch
most dangerous quakes without too many false alarms. The issue is cut and dried:
if useful precursors exist, earthquake prediction is possible; if they don’t,
it’s not.
Twenty-five years ago, scientists in the US thought they were hot on the
trail of such precursors. They looked for them in patterns of small earthquakes,
movements of the Earth’s crust, electromagnetic signals emitted by rock being
squashed or split by tectonic forces, changes in level of water in wells and in
the amount of radon gas seeping from the ground. The research went into
overdrive when scientists theorised that several promising candidates might stem
from a common cause. Lab results showed that when rock is stressed to near its
breaking point, it becomes shot through with a network of microscopic fractures
that fluff it up like a pillow. That expansion—called
dilatancy—along with the subsequent draining and refilling of groundwater
in the cracks, looked as if it could explain a host of changes in electrical
conductivity, the speed of seismic pressure waves, even the elevation of the
ground in earthquake-prone areas. If a grand unified theory of earthquake
precursors were close at hand, could prediction itself be far behind?
Overseas, it looked as though prediction might already be becoming a reality.
In 1975, after an unusual swarm of small quakes rattled Haicheng in Manchuria,
China, the city authorities alerted residents in time to save lives. Some 90 per
cent of the city’s buildings were destroyed or damaged, but casualties were low.
Even without a grand unified theory, it looked as if prediction was working.
Then the bubble burst. When scientists took measurements in the field, they
found that dilatancy-diffusion applied only to small portions of the
crust—too small to explain the precursors. Worse, when they double-checked
the precursors themselves, many of them proved to be measurement errors or
wishful thinking. In China, the success at Haicheng turned out to be a fluke.
The following year, an unpredicted earthquake levelled the city of Tangshan in
Shandong just 450 kilometres away and killed at least 250 000 of its million
inhabitants. To this day, scientists disagree about whether the Haicheng
prediction was more than a lucky guess.
Searching for still subtler signals, scientists in the US decided to
concentrate their efforts on one big experiment: a patch of ranchland bristling
with instruments near the hamlet of Parkfield in central California, where
earthquakes measuring 6 on the Richter scale seemed to strike every 22 years
with near-clockwork regularity. By watching carefully for precursors, they hoped
to predict the next quake, which experts estimated had even odds of striking by
1988 and a 95 per cent chance of striking by 1993 (see Diagram).
Fourteen years after the Parkfield Earthquake Prediction Experiment began,
the expected earthquake still has not come. Elsewhere in the US, prediction
research has ground to a virtual standstill, a casualty, its proponents say, of
the field’s unfulfilled early promise. “We were very naive,” says Max Wyss,
chairman of the earthquake prediction subcommittee of the International
Association of Seismology and Physics of the Earth’s Interior. “Much of
earthquake prediction research was done very poorly.”
The main problem, says David Jackson, science director of the Southern
California Earthquake Center in Los Angeles and a coauthor of Geller’s article
in Science, was the philosophy that you didn’t need to understand very
much about earthquakes to predict them. That black-box approach to science has
been largely abandoned in the US, but it still flourishes in Robert Geller’s
adopted country, Japan. And, say Western critics, all Japanese prediction
researchers have to show for their ever-increasing budget (worth $185
million last year) is a record of unrelenting failure.
Most quake prediction scientists in the US read Geller’s broadsides as coded
attacks on Japanese research priorities—attacks with which, by and large,
they agree. But in the US, they say, the target of the criticism is hardly worth
hitting. At the US Geological Survey, shrunk by downsizing and straitened by
policies favouring research with immediate practical applications, prediction is
distinctly démodé. “It’s a bad word,” says John Langbein,
chief scientist for the Parkfield experiment (whose operating cost of $1
million per year makes up only about 2 per cent of the USGS’s budget for
earthquake research). It is much the same all over, says long-time prediction
enthusiast Paul Silver, a seismologist at the Carnegie Institution of
Washington. “Nobody likes to be unsuccessful, so there’s been a shift in
emphasis toward things we know how to do.”
These things include long-term forecasts of quake probabilities, studies of
how the ground will shake during future earthquakes, and warning systems to
detect major quakes as they arise. One such system, a network of sensors known
as TriNet, is being installed in Southern California. The five-year, $20
million project will help emergency managers assess damage just minutes after an
earthquake strikes. Future versions may sound an alarm in the seconds while
seismic waves are still on the way. The new watchword is hazard assessment:
finding where the danger lies and building cities to survive quakes, whenever
they choose to strike.
Some researchers in the US, however, refuse to give up on prediction. The
black-box approach has not worked, they say. But that doesn’t mean that you
should abandon the whole idea. “The emphasis from the beginning should have been
to understand the processes that accompany the occurrence of earthquakes,” says
Allan Lindh, a USGS geophysicist who led the lobbying campaign to set up the
Parkfield experiment. “When you understand things well enough, then you start to
get better at predicting them.”
The problem is that quakes are fiendishly hard to study. For a start, Lindh
says, “there are 10 to 15 kilometres of dirt, rock and fluids between us and
where the action is.” Scientists can study that action indirectly by measuring
seismic waves, electromagnetism and strain—the movement and deformation of
rock under tectonic stresses—but their remote readings, distorted by
intervening layers, are a poor substitute for being there. Furthermore, big
earthquakes, which are the ones of most interest to forecasters, are rare and
strike capriciously in space and time. And everything happens fast. The rupture
begins, or nucleates, in a small patch of fault and spreads at speeds of
kilometres per second—too fast for current instruments to pick up the
details. The longer the rupture, the bigger the earthquake. But are nucleation
zones too small to be detected? Are they all the same size, or do they vary? Is
the size related to the length of the rupture? No one knows.
Impossible reality
Faced with these and other difficulties, some critics argue that prediction
is out of the question. “I actually think that it’s impossible in principle,”
says Lucile Jones, a USGS seismologist working on the TriNet project. The
problem, she says, is that rather than predicting every quake, you need to
predict which of the many that occur every day will turn into large, damaging
ones. The magnitude of an earthquake, she points out, depends on the length of
the rupture. That, in turn, depends not on where the rupture starts, but where
it stops. And where it stops, some earthquake scientists believe, can be just
about anywhere. A rupture is not a single event, but a series of tiny
ones—a cascade that gathers momentum as it goes, like a landslide.
According to some models, a rupturing fault resembles a pile of sand so
delicately poised that there is no way to tell whether the next grain you add
will drop quietly into place or trigger a landslide.
In chaos theory, that kind of delicate balance is called “self-organised
criticality”. What it implies for earthquake prediction, opponents say, is dire:
there is no way to tell in advance whether a rupture will be short or long, and
thus there is no way of predicting whether the earthquake will be imperceptible
or colossal. “If that’s true,” says Jones, “earthquake prediction as people want
it is impossible.”
Open question
That’s a big “if”. For one thing, chaos alone does not necessarily imply
unpredictability. “Planetary orbits are chaotic, but that doesn’t mean I can’t
predict the orbit of Mars,” says Lindh. According to Charles Sammis, a
geophysicist at the University of California, Los Angeles, who is working on
chaos-based methods for analysing earthquake statistics, prediction is
impossible only if all of the Earth’s crust is in a state of self-organised
criticality all the time. But that is as unrealistic as assuming that the
Earth’s surface has a uniform temperature. Sammis says the criticality
conditions vary widely from place to place and fluctuate over time. A big
earthquake, for example, can render a whole region temporarily
non-critical—just as a big landslide can flatten out the lower slope of a
pile of sand. The challenge is to identify where earthquakes are predictable at
any one moment. That, he says, is an open question.
It’s one thing to show that quake prediction is “not impossible”, and quite
another to come up with a reliable precursor. But tantalising evidence of at
least one such predictor has already turned up in the laboratory. If you
simulate a fault by sawing a chunk of granite in half and then squeezing the
halves back together with a hydraulic press, slightly before the miniature fault
ruptures, nearby rock begins to shift almost imperceptibly. James Dieterich, now
director of earthquake hazards at USGS’s Western Region headquarters in Menlo
Park, California, described this “premonitory creep” mathematically in the
early 1980s. Along real faults, he says, such slip might conceivably warn of
impending earthquakes if it is large enough to be detected. Unfortunately,
Dieterich’s experiments leave this question open.
In 1997, seismologists at the USGS and at Stanford University in California
found subtle evidence of such creep in seismic data from earthquakes around the
world. But their observations might also have resulted from a landslide-like
cascade of tiny ruptures. Another candidate appeared in August, when creepmeters
near San Juan Bautista, California detected an unusual creep of a fraction of a
millimetre directly over the spot where, a week later, a quake nucleated that
reached 5.1 on the Richter scale. Dieterich won’t swear that it was premonitory
creep, but says it’s the best documented case yet seen.
Whether or not premonitory creep is the key to earthquake prediction, many
scientists believe that any precursors that do turn up will have at least two
things in common with it: they will be furtive— researchers have long
since refuted all the blatant ones—and they will appear as anomalies in
measurements of strain. That’s only natural, Silver says: most would-be
precursors investigated so far, including changes in well levels and in
groundwater geochemistry have been by-products of compressed rock.
So instead of investigating a ragbag of indicators, it makes sense to measure
strain directly. The trouble is that existing receiver networks for the Global
Positioning System of satellites are not sensitive enough. The GPS is perfect
for tracking the slow, steady accumulation of strain that raises mountains and
provides a basis for 30-year earthquake forecasts. Individual earthquakes,
however, are more likely to be heralded by much shorter changes in strain,
lasting hours or weeks. Such short-term changes, or transients, slip right past
the GPS but can be detected by strainmeters—underground barometers capable
of sensing a change in rock pressure of one part per billion. Strainmeters are
expensive to install so they tend to be scarce. According to Paul Silver,
Southern California will soon boast hundreds of GPS receivers, but it has only
five strainmeters.
Silver would like to remedy that. In the March/April 1998 issue of
Seismological Research Letters he proposed a network of up to 2000 new
monitoring stations to record seismic waves, strain, strain transients and
ground shaking along the Pacific coast. The “plate boundary deformation network”
would run for about 20 years at a cost of $25 million a year. By the end,
he says, scientists would have racked up data from enough quakes to know for
sure whether they are predictable.
More and better measurements are all very well, says Jackson, but even if the
precursors exist, it would take a century or more to log enough big earthquakes
to find out how they work. That doesn’t worry prediction advocates. They have
long realised that effective tools for prediction will not arrive overnight, but
they still insist it should be done.
Think of cancer research, says geophysicist Mark Zoback of Stanford
University. Finding a cure is clearly a very hard problem, but that’s not to say
all effort should be stopped. “How would the public feel if we abandoned the
quest for a cure and just issued pronouncements on health like try not to smoke
and stay out of the sun?”
If we give up on prediction research, he says, we’re in danger of violating
the public’s trust. “We shouldn’t promise something that we can’t deliver, and
if in the end quakes are not predictable, so be it. But it’s too easy to work on
problems that can be solved rather than ones that should be solved. This is
stuff that’s worth doing, and worth doing well.”
