can be made. Short-term predictions are made using various
phenomena that precede major eruptions: rising magma causes
ground deformation, small earthquakes, and the release of
volcanic gases. Finally, small eruptions usually precede a large
one, making it possible to issue real-time warnings. Hence the
record of volcanic predictions, though not perfect, 8 is reasonably good. The area around Mt St Helens was evacuated before
the giant eruption of May 18, 1980, reducing the loss of life
to only 60 people, including a geologist studying the volcano
and citizens who refused to leave. The largest eruption of
the second half of the twentieth century, Mt Pinatubo in the
Philippines, destroyed over 100,000 houses and a nearby US
Air Force base, yet only 281 people died because of evacuations
during the preceding days.
Seismologists would like to do as well for earthquakes. We
would like to be able to forecast where they are on average
likely to occur in years to come, predict them a few years to
hours before they occur, and issue real-time warnings after an
earthquake has occurred in situations where such a warning
would be useful. However, the record of seismology in these
areas is mixed. To date there has been some success in longterm forecasting, little if any in short-term prediction, and
some in real-time warning.
Earthquake forecasting, discussed in Section 4.7.3, estimates
the probability that an earthquake of a certain magnitude will
occur in a particular area during a specific time. For instance,
a forecast might be a 25% probability of a magnitude 7 or
greater earthquake occurring along the San Francisco segment
of the San Andreas fault in the next 30 years. Forecasting uses
the history of earthquakes on the fault and other geophysical
information, such as the crustal motions measured using the
Global Positioning System, to predict its likely future behavior.
While forecasting is not relevant to short-term earthquake
preparations, it is important in the enactment of building codes
for earthquake-resistant construction, which are costly and
require justification. Such forecasting is already successful in
general ways; knowing that the San Andreas and nearby faults
will be the sites of recurrent earthquakes has prompted building codes that are a major reason why the 1989 Loma Prieta
and 1994 Northridge earthquakes caused few casualties.
Going beyond general forecasts is more difficult. For example, the probabilistic hazard map for the USA in Fig. 1.2-3
predicts a general pattern of higher hazards in areas of known
past large earthquakes. Most of these, in California and
Nevada, the Pacific Northwest, and Utah, are in the western
USA, in the broad boundary zone between the Pacific and
North American plates. In addition, high hazards are predicated in parts of the interior of the continent, near Charleston,
South Carolina, and the New Madrid seismic zone in the
Midwest. The map attempts to quantify this risk in terms of the
maximum expected acceleration (recall that 0.2 g corresponds
approximately to the onset of significant building damage)
during a time interval. Such maps are made by assuming where
and how often earthquakes will occur, how large they will be,
and then using ground motion models like those in Fig. 1.2-5 to
predict how much ground motion they will produce. Because
these factors are not well understood, especially in intraplate
regions where large earthquakes are rare, hazard estimates
have considerable uncertainties. 9 For example, the high hazard
predicted for parts of the Midwest, exceeding that in San
Francisco or Los Angeles, results from specific assumptions,
and alternative assumptions yield quite different estimates
(Fig. 1.2-14).
Similarly, hazard estimates depend on the probability and
hence recurrence time considered. Where the largest earthquakes are expected about every 200 years a for example, near
a plate boundary as in California a a hazard map predicting
the maximum acceleration expected at a 10% probability in the next 50 years, or at least once during the next 500
(50/0.1) years, will be similar to one for 2% probability in the
next 50 years, or at least once during the next 2500 (50/0.02)
years, because each portion of plate boundary is expected to
rupture at least once in 500 years. However, the two maps
would differ significantly where large earthquakes are less
frequent a for example, in an intraplate region like the New
Madrid zone (Sections 4.7.1, 5.6.3). This issue is important in
choosing building codes because typical buildings have a useful
life of about 50 years.
Because earthquakes are infrequent on a human time scale,
it will be a long time before we know how well such estimates,
which combine long-term earthquake forecasts and ground
motion predictions, actually describe future earthquakes.
Nonetheless, such estimates are used for purposes such as
developing building codes and setting insurance rates. As a
result, how to make meaningful predictions and hazard estimates, communicate their uncertainties to the public, and best
use them for policy is a topic of discussion relevant not just
to seismology but to the other earth sciences as well.
A key scientific challenge for hazard estimation is that the
process determining when large earthquakes recur is unclear.
The underlying basis for seismic forecasting is the principle of
elastic rebound (Section 4.1). In this model, large-scale crustal
motions, in most cases due to plate motions, slowly build up
stress and strain across locked faults. When the stress reaches
a critical threshold, seismic slip occurs along the fault, and the
stress immediately drops. The process then begins again. The
repeat time for these earthquakes depends on the rate at which
crustal motions load the fault and the properties of the rocks
that control when it slips.
8 In 1982, uplift of the volcanic dome and other activity near the resort town of
Mammoth Lakes, California, suggested that an eruption might be imminent. Geologists issued a volcano alert, resulting in significant tensions with local business leaders.
When no eruption occurred, geologists were the target of much local anger, and the
county supervisor who arranged for an escape route in the event of a volcanic eruption
was recalled in a special election.
9 Earthquake risk assessment has been described as “a game of chance of which we
still don’t know all the rules” (Lomnitz, 1989).
1.2 Seismology and society 21
phenomena that precede major eruptions: rising magma causes
ground deformation, small earthquakes, and the release of
volcanic gases. Finally, small eruptions usually precede a large
one, making it possible to issue real-time warnings. Hence the
record of volcanic predictions, though not perfect, 8 is reasonably good. The area around Mt St Helens was evacuated before
the giant eruption of May 18, 1980, reducing the loss of life
to only 60 people, including a geologist studying the volcano
and citizens who refused to leave. The largest eruption of
the second half of the twentieth century, Mt Pinatubo in the
Philippines, destroyed over 100,000 houses and a nearby US
Air Force base, yet only 281 people died because of evacuations
during the preceding days.
Seismologists would like to do as well for earthquakes. We
would like to be able to forecast where they are on average
likely to occur in years to come, predict them a few years to
hours before they occur, and issue real-time warnings after an
earthquake has occurred in situations where such a warning
would be useful. However, the record of seismology in these
areas is mixed. To date there has been some success in longterm forecasting, little if any in short-term prediction, and
some in real-time warning.
Earthquake forecasting, discussed in Section 4.7.3, estimates
the probability that an earthquake of a certain magnitude will
occur in a particular area during a specific time. For instance,
a forecast might be a 25% probability of a magnitude 7 or
greater earthquake occurring along the San Francisco segment
of the San Andreas fault in the next 30 years. Forecasting uses
the history of earthquakes on the fault and other geophysical
information, such as the crustal motions measured using the
Global Positioning System, to predict its likely future behavior.
While forecasting is not relevant to short-term earthquake
preparations, it is important in the enactment of building codes
for earthquake-resistant construction, which are costly and
require justification. Such forecasting is already successful in
general ways; knowing that the San Andreas and nearby faults
will be the sites of recurrent earthquakes has prompted building codes that are a major reason why the 1989 Loma Prieta
and 1994 Northridge earthquakes caused few casualties.
Going beyond general forecasts is more difficult. For example, the probabilistic hazard map for the USA in Fig. 1.2-3
predicts a general pattern of higher hazards in areas of known
past large earthquakes. Most of these, in California and
Nevada, the Pacific Northwest, and Utah, are in the western
USA, in the broad boundary zone between the Pacific and
North American plates. In addition, high hazards are predicated in parts of the interior of the continent, near Charleston,
South Carolina, and the New Madrid seismic zone in the
Midwest. The map attempts to quantify this risk in terms of the
maximum expected acceleration (recall that 0.2 g corresponds
approximately to the onset of significant building damage)
during a time interval. Such maps are made by assuming where
and how often earthquakes will occur, how large they will be,
and then using ground motion models like those in Fig. 1.2-5 to
predict how much ground motion they will produce. Because
these factors are not well understood, especially in intraplate
regions where large earthquakes are rare, hazard estimates
have considerable uncertainties. 9 For example, the high hazard
predicted for parts of the Midwest, exceeding that in San
Francisco or Los Angeles, results from specific assumptions,
and alternative assumptions yield quite different estimates
(Fig. 1.2-14).
Similarly, hazard estimates depend on the probability and
hence recurrence time considered. Where the largest earthquakes are expected about every 200 years a for example, near
a plate boundary as in California a a hazard map predicting
the maximum acceleration expected at a 10% probability in the next 50 years, or at least once during the next 500
(50/0.1) years, will be similar to one for 2% probability in the
next 50 years, or at least once during the next 2500 (50/0.02)
years, because each portion of plate boundary is expected to
rupture at least once in 500 years. However, the two maps
would differ significantly where large earthquakes are less
frequent a for example, in an intraplate region like the New
Madrid zone (Sections 4.7.1, 5.6.3). This issue is important in
choosing building codes because typical buildings have a useful
life of about 50 years.
Because earthquakes are infrequent on a human time scale,
it will be a long time before we know how well such estimates,
which combine long-term earthquake forecasts and ground
motion predictions, actually describe future earthquakes.
Nonetheless, such estimates are used for purposes such as
developing building codes and setting insurance rates. As a
result, how to make meaningful predictions and hazard estimates, communicate their uncertainties to the public, and best
use them for policy is a topic of discussion relevant not just
to seismology but to the other earth sciences as well.
A key scientific challenge for hazard estimation is that the
process determining when large earthquakes recur is unclear.
The underlying basis for seismic forecasting is the principle of
elastic rebound (Section 4.1). In this model, large-scale crustal
motions, in most cases due to plate motions, slowly build up
stress and strain across locked faults. When the stress reaches
a critical threshold, seismic slip occurs along the fault, and the
stress immediately drops. The process then begins again. The
repeat time for these earthquakes depends on the rate at which
crustal motions load the fault and the properties of the rocks
that control when it slips.
8 In 1982, uplift of the volcanic dome and other activity near the resort town of
Mammoth Lakes, California, suggested that an eruption might be imminent. Geologists issued a volcano alert, resulting in significant tensions with local business leaders.
When no eruption occurred, geologists were the target of much local anger, and the
county supervisor who arranged for an escape route in the event of a volcanic eruption
was recalled in a special election.
9 Earthquake risk assessment has been described as “a game of chance of which we
still don’t know all the rules” (Lomnitz, 1989).
1.2 Seismology and society 21