Sie sind auf Seite 1von 16

International Journal of Disaster Risk Reduction 1 (2012) 17–32

Contents lists available at SciVerse ScienceDirect

International Journal of Disaster Risk Reduction


journal homepage: www.elsevier.com/locate/ijdrr

Advance prediction of the March 11, 2011 Great East Japan


Earthquake: A missed opportunity for disaster preparedness
C. Davis a,n, V. Keilis-Borok b,c, V. Kossobokov c,d, A. Soloviev c
a
Geotechnical Engineering Group, Los Angeles Department of Water and Power, 111 North Hope Street, Room 1368, Los Angeles, CA 90012, USA
b
Department of Earth and Planetary Sciences and Institute of Geophysics and Planetary Physics, University of California, Los Angeles, CA 90095-1567, USA
c
Institute of Earthquake Prediction Theory and Mathematical Geophysics, Russian Academy of Sciences, 84/32 Profsouznaya, Moscow 117997, Russia
d
Institut de Physique du Globe de Paris, 1 Rue Jussieu, 75238 Paris Cedex 05, France

a r t i c l e i n f o abstract

Article history: There was a missed opportunity for implementing important disaster preparedness
Received 5 February 2012 measures following an earthquake prediction that was announced as an alarm in mid-
Received in revised form 2001. This intermediate-term middle-range prediction was the initiation of a chain of
28 February 2012
alarms that successfully detected the time, region, and magnitude range for the
Accepted 8 March 2012
Available online 6 April 2012
magnitude 9.0 March 11, 2011 Great East Japan Earthquake. The prediction chains
were made using an algorithm called M8 and is the latest of many predictions tested
Keywords: worldwide for more than 25 years, the results of which show at least a 70% success rate.
Earthquake prediction The earthquake detection could have been utilized to implement measures and improve
Forecast
earthquake preparedness in advance; unfortunately this was not done, in part due to
Emergency management
the predictions’ limited distribution and the lack of applying existing methods for using
Disaster preparedness
Great East Japan Earthquake and tsunami intermediate-term predictions to make decisions for taking action. The resulting
earthquake and induced tsunami caused tremendous devastation to north-east Japan.
Methods that were known in advance of the predication and further advanced during
the prediction timeframe are presented in a scenario describing some possibilities on
how the 2001 prediction may have been utilized to reduce significant damage,
including damage to the Fukushima nuclear power plant, and to show prudent cost-
effective actions can be taken if the prediction certainty is known, but not necessarily
high. The purpose of this paper is to show how the prediction information can be
strategically used to enhance disaster preparedness and reduce future impacts from the
world’s largest earthquakes.
& 2012 Elsevier Ltd. All rights reserved.

1. Introduction the Tohoku Earthquake) was detected years in advance


using a combined earthquake prediction algorithm
The world’s largest earthquakes are predictable within called M8–MSc, which is based on premonitory seismicity
time, space, and magnitude ranges having limited but patterns and prior to this prediction had been validated
known accuracy. The March 11, 2011 Great East Japan by predictions-in-advance. The prediction was initially
Earthquake, off shore of the Tohoku region, (herein called announced in mid-2001 nearly ten years in advance of the
earthquake using a standard protocol to a restricted group
of experts and identified the possibility of a magnitude
n
Corresponding author. Tel.: þ1 213 367 0855; (M) 8.0 or greater earthquake occurring in an area
fax: þ 1 213 367 3792.
covering northern Honshu and Hokkaido. The M8 earth-
E-mail addresses: craig.davis@ladwp.com (C. Davis),
vkb@ess.ucla.edu (V. Keilis-Borok), volodya@mitp.ru (V. Kossobokov), quake prediction algorithm has at least a 70% success rate
soloviev@mitp.ru (A. Soloviev). of importance to disaster managers for predicting great

2212-4209/$ - see front matter & 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.ijdrr.2012.03.001
18 C. Davis et al. / International Journal of Disaster Risk Reduction 1 (2012) 17–32

(M8.0þ) earthquakes around the world. Since 1992 M8 has the devastating mega-earthquake that struck Japan on
predicted 12 out of 17 earthquakes for this magnitude range March 11, 2011. This was a missed opportunity that can
including the 2010 Chili earthquake. The place and time for be rectified in future earthquakes by establishing better
each of three mega-earthquakes (2004 Indian Ocean, 2010 links and lines of communication between the earthquake
Chili, and 2011 Japan) were recognized as in state of prediction and disaster preparedness communities.
increased probability by the same algorithm aimed at larger Intermediate-term predictions are not able to be
magnitude ranges of mega-earthquakes (i.e., M8.5þ and utilized to prevent all damage and save all human lives,
M9.0þ ), though in the case of the 2004 Indian Ocean mega- but they can be used to take certain cost effective actions
thrust it was a retrospective application [34]. to reduce damage and casualties and improve post-
The 2001 prediction and subsequent extensions pro- disaster recovery. Molchan [39,40] and Davis et al. [7]
vided a (missed) possibility for considerable damage present methodologies and examples on how actions may
reduction. This is a part of the much wider field: chaos be taken in response to predictions. Davis [5] describes
theory/statistical physics approach to predicting extreme how to apply economic parameters to loss equations for
events; joint optimization of prediction and preparedness use in optimizing actions that may be taken in response to
[30,14]. a prediction. Such actions are useful to enhance the
An earthquake prediction includes statements for normal long-term seismic risk mitigation methods like
range of magnitude, geographical area (location), time building codes and standard disaster preparedness meth-
interval, and probability of a future event [1]. Currently ods. A simple description is provided herein to show some
there are no identified reliable methods that can precisely possibilities on how the 2001 prediction may have been
predict the location of a damaging earthquake within an utilized to reduce significant damage to the Fukushima
immediate timeframe (i.e., a few hours to days). Complex nuclear power plant and other infrastructure; other cost-
systems such as earthquakes generated in the lithosphere effective actions such as increasing inventory and aspects
are not predictable with an absolute precision. A danger- for business continuity following an event could be
ous mistake sometimes emerges in literature stating only implemented but are not described herein. This is done
precise predictions in space, time, and magnitude are only with the intent to develop a lesson we can all learn
useful. Actually, predictions might be useful if their from in order to help reduce seismic risks. This study
accuracies are known, but not necessarily high. Such is identifies the importance of using credible earthquake
the standard practice to all prediction problems, including predictions to help reduce future potential damage from
national defense. The M8 algorithm predicts earthquakes the world’s largest earthquakes.
with intermediate-term (several years) middle-range
(areas of linear dimension 5 to 10 times the earthquake 2. Released-in-advance prediction: March 11, 2011 M 9.0
source zone size) accuracy. The MSc algorithm decreases Tohoku Japan Earthquake
the prediction area to a narrow range (2–3 times the
earthquake source zone size) and in some cases nearly an Since the design of the M8 algorithm [26] it was used for
exact location (within the dimension of the source zone). the prediction of earthquakes about magnitude 8.0 (hence, its
The Tohoku earthquake and tsunami created a great name) then applied in other magnitude ranges down to
disaster. On March 11, 2011 Japan was subjected to the magnitude 5 [11,32,34–36,45]. The M8 and MSc earthquake
largest known shaking from one of the world’s greatest prediction algorithms [28,37] had been used in combination
earthquakes, followed by one of the world’s largest worldwide in advance of the 1988 Spitak (Armenia) and
tsunamis, and the initiation of the world’s second largest 1989 Loma Prieta (California) earthquakes, whose occur-
nuclear accident. This earthquake directly affected nearly rences did confirm the M8–MSc predictions. However, the
41 million people with very strong to extreme shaking. official systematic continuous global test was launched in
There were 15,844 people killed and 3450 people remain 1992 [18] as a result of the Russian–American collaboration.
missing [43]. Millions of people were left without water The M8–MSc prediction identifies a Time of Increased Prob-
and power. Entire towns were reduced to rubble. Over ability (TIP) of an earthquake with magnitude M from a given
440,000 people have been displaced from their homes by magnitude range MM0 þ ¼[M0, M0 þ DM] occurring within a
the earthquake, tsunami, fire, or nuclear radiation alert defined area. (The M8 and MSc algorithms are described in
and radioactive contamination of the environment. Losses Appendix A.)
may reach or exceed $235 billion and insurers may pay In essence, the prediction methodology of M8 and MSc
out over $42 billion, making this the world’s costliest algorithms integrated expertise in statistical modeling of
natural disaster [46–48,50]. complex systems, non-linear physics of the solid Earth,
Several decades ago pattern recognition methods [12,13] and exploratory data analysis; the framework for the
were used to identify how the same area in Japan is integration was provided by pattern recognition of infre-
susceptible to earthquakes of magnitude 8.2 or greater quent events (Appendix B describes more detail).
[16,17] and paleo-tsunami investigations identified magni- Fig. 1 and Table 1 summarize the global testing of the
tude 8.3 earthquakes were possible [38]. This is in contrast M8–MSc algorithm in the areas from Kamchatka to the
to the regional magnitude 7.4–7.7 earthquake hazards that Marianas, including the Japanese islands, for the timeframe
were used to prepare the affected communities for earth- 1985–2011. Within this time and space alarms were called
quake strikes [3]. Unfortunately, the M8–MSc prediction, targeting at least one earthquake from M8.0þ . All alarms
pattern recognition, or paleo-tsunami information appar- were initially called for 5 years, covered ‘‘circular’’ areas of
ently were not used to make the necessary preparations for about 1,400,000 km2, had a probability of false alarm f not
C. Davis et al. / International Journal of Disaster Risk Reduction 1 (2012) 17–32 19

Fig. 1. Global testing of algorithms M8 and MSc aimed at magnitude range M Z 8.0: Space–time distribution of alarms from Kamchatka to the Marianas.
Circles of Investigation (CI, white light) are shown on the left; the space–time distribution of alarms from 1985 to 2011 is on the right (dark gray for M8
and black for MSc). The space coordinate is the distance along the belt (six arrows point to the centers of corresponding circles of investigation). Great
earthquakes are marked by stars. The 2011 mega-thrust (‘‘2011 off the Pacific coast of Tohoku Earthquake’’—JMA) is indicated by the big star. The M8
alarm had a July 1, 2011 tentative expiration date but was terminated by margin on January 1, 2011; marked with light gray on the space–time plot.

Table 1 hand side of Fig. 1 for identification of the alarms and their
Prediction targets: Magnitude M8.0 þ earthquakes in Kamchatka– duration.
Japan–Marianas, 1993–2011—data available at http://www.mitp.ru/en/
predictions.html.
All alarms in Fig. 1 were announced in advance and
posted at the following web sites:
Date (yr/mo/day) Lat Lon Depth M Result
(km)  http://www.mitp.ru/predictions.html.
1993/08/08 12.98 144.80 59 8.2 Predicted
 http://users.ictp.it/www_users/sand/index_files/
1994/10/04 43.77 147.32 14 8.1 Predicted DevelopmentofPrediction.html.
1995/12/03 44.66 149.30 33 8.0 Predicted
2003/09/25 41.81 143.91 27 8.3 Missed
2006/11/15 46.59 153.27 10 8.3 Missed As is done with all yet unexpired alarms, access to the
2007/01/13 46.24 154.52 10 8.2 Missed
2011/03/11 38.32 142.37 24 9.0 Near miss/
alarms was limited to about 200 scientists and profes-
outside M8.0 þ sional experts worldwide. This is done in compliance to
the United Nations Education, Scientific, and Cultural
Organization (UNESCO) guidelines [10] and the Seismo-
greater than 30%, and were re-evaluated every six months. logical Society of America Guidelines for Earthquake
Each re-evaluation resulted in either continuing the alarm Predictors [53] for the public release of prediction,
as originally called, changing the M8 and/or MSc alarm area, because it might trigger social disruption, anxiety of
canceling the alarm, or extending the alarm duration. For population, profiteering, crime, etc. The predictions are
purposes of this paper, and for emergency management use, made available to the general public after a strong earth-
the total alarm region is defined as the seismic locus within quake occurs or the alarm expires, whichever comes first.
the sum of all adjoining circular areas with TIP at the same One alarm sequence of particular importance is that
time, and the total alarm duration is the total time con- shown on the far right of Fig. 1. This alarm initiated in
tinuum for the alarm region. Thus, for this purpose Fig. 1 the middle of 2001 and lasted until 2011. Within this
shows four alarms (these definitions are technically differ- sequence the last earthquake prediction update made by
ent than those used to call individual alarms in each circle of the combined M8–MSc algorithms was released in July
investigation as described in Appendix A. The above defini- 2010, and covered the time until July 2011. This predic-
tions are useful for applying the alarm for emergency tion identified the possibility of at least one earthquake of
management and disaster preparedness purposes). The magnitude 8.0 or greater in Japan prior to March 11, 2011
projection of the prediction regions on the seismic belt axis within the Circle of Investigation (CI) #81 area shown in
at a given time within 1985–2011 are plotted on the right Fig. 1. Fig. 2 identifies the approximate alarm regions.
20 C. Davis et al. / International Journal of Disaster Risk Reduction 1 (2012) 17–32

MSc algorithms that has been on-going since 1992 and


Prediction Regions (approximate)
therefore cannot be formally regarded as documentation in
M8 2001-2006
M8 2006-2011
advance of the March 11, 2011 earthquake as part of the
MSc 2001-2006 global testing for M8.0þ earthquakes. Even so, from a
MSc 2006-2011 disaster preparedness perspective the predictions could
have been useful for improving seismic risk management
if the information was utilized to implement appropriate
measures. Unfortunately, the Japanese experts were caught
off guard by the Tohoku earthquake, in part because the
disaster managers were not properly linked with the earth-
quake prediction community resulting in inadequate com-
munication and lack of use of important information.
For the region presented in Fig. 1 seven target earth-
quakes occurred between 1993 and 2011 and four alarms
were called, two of them false. Table 1 shows that three
target earthquakes were predicted and four were missed,
Fig. 2. Schematic illustrating the approximate M8 and MSc alarm with the March 11, 2011 Tohoku earthquake being
regions called during 2001–2006 and 2006–2011 in relation to the identified as a near miss as previously described. The
Japanese Islands and the tectonic plate boundaries. The star identifies most recent chain of alarms had earthquakes occur short
approximate epicenter for the March 11, 2011 Tohoku earthquake. Base
distances outside of the defined prediction regions, an
drawing from NIED [44]. Exact locations for circles of investigation are
found at http://www.mitp.ru/en/predictions.html. aspect important to emergency managers in relation to
the need for preparing. This chain is technically consid-
ered a false alarm and the missed earthquakes are
The large circular area, shown in light gray in Fig. 1, is classified as failures to predict because the algorithms
predicted by the M8 algorithm. The MSc algorithm was did not detect those specific areas as having a TIP. None-
used to refine the prediction areas to within the smaller theless, the algorithm did detect regional characteristics
rectangular areas in Fig. 2, shown as black in Fig. 1. As for potential great earthquakes which could have been
seen in Figs. 1 and 2, this alarm covered the area where used to enhance preparedness in areas damaged by some
the Tohoku Japan mega-earthquake actually occurred, of the undetected earthquakes in Table 1. For example, at
and the epicenter is located within the refined area the end of the following section a scenario description is
predicted by the MSc algorithm. The MSc prediction provided on how the 2001 prediction could have aided in
region from 2001 to 2011 is located in the Tohoku Region reducing damage in Hokkaido as a result of the 2003 M
near the east coast of Honshu, Japan. 8.3 Tokachi-Oki earthquake.

2.1. Review of M8–MSc advanced prediction


3. Missed opportunity
In mid-2001, the combined M8–MSc algorithm recog-
nized premonitory patterns and diagnosed a TIP for a An opportunity to reduce the impacts from the Tohoku
MZ8.0 earthquake in the region shown in Figs. 1 and 2, earthquake and tsunami disaster has been missed. The
and an alarm was announced. Using standard procedure for earthquake predictions shown in Fig. 1 were not used to
evaluating the algorithm, updates were made every 6 improve disaster preparedness in the regions of greatest
months until January 2011. In January 2011 the algorithm risk. Fig. 3 shows a simple and straight forward process
for predicting M8.0þ earthquakes detected a small change where the information provided by credible earthquake
in one of the characteristics (the concentration of sources), predictions can be used to get properly prepared for the
which placed its value just below the criteria necessary to predicted hazard; a simplified diagram of common seis-
maintain the alarm (see Appendix A for more details on mic risk reduction practices. Davis et al. [6,7] provide
alarm diagnosis). This small change in value resulted in the examples on how predictions such as that described
alarm being called off on January 1, 2011, only 70 days herein can be cost effectively used by infrastructure
before the M 9.0 Tohoku event [34]. As shown in Figs. 1 and owners, disaster managers, and policy makers for redu-
2, the premonitory detection and application of the M8–MSc cing seismic risk. A general methodology for such esti-
algorithms accurately identified the Tohoku earthquake. The mates based on optimal control theory is presented by
epicenter was located clearly within the territory defined by Molchan [39,40]. Some of their examples and scenarios
the MSc algorithm, showing how it successfully identified could have been considered in advance by the Japanese.
the narrow range within which the earthquake occurred. The specific actions that may be implemented in any
However, the Tohoku earthquake will not be counted as a region in response to any prediction requires knowledge
successful prediction by the M8 algorithm in the magnitude of the local conditions at the time of prediction and an
range M8.0þ. A concurrent alarm from a M8 algorithm understanding of resident cultures and values. This
aimed to predict M8.5þ mega-earthquakes remained in undoubtedly holds true for Japan prior to the Tohoku
place at the time the M9.0 mega-thrust earthquake occurred earthquake. However, there are also many universal
in Japan [34]; but the use of this algorithm for M8.5þ and actions of great importance which are described in the
M9.0þ is beyond the scope of the Global test of M8 and following subsection.
C. Davis et al. / International Journal of Disaster Risk Reduction 1 (2012) 17–32 21

Fig. 3. Process for determining actions to implement in response to an earthquake prediction. Smaller bold rectangular boxes provide specific
information related to the larger boxes and as described in the text.

3.1. Scenario use of prediction 50% within the smaller MSc rectangular grids. These prob-
abilities are based on global statistics (e.g. [34]), not the
In order to provide better insight into how the predictions regional data shown in Table 1. This prediction is assessed as
could have been utilized, the following scenario is described. credible based on the scientific manner in which the algo-
The presented scenario is simplified and not intended to rithms have been developed, their performance over the past
trivialize the necessary scientific or managerial efforts or 10 years, and recent successful predictions in advance
attempt to cover the potentially vast and laborious under- including the August 8, 1993 M 8.0 Guam, October 4, 1994
takings that could accompany a true response, and it has the M 8.3 Shikotan, April 7, 1995 M 8.1 Samoa, December 3, 1995
advantage of hindsight and knowledge of what did transpire M 8.0 Iturup, February 17, 1996 M 8.2 New Giunea, and June
after the Tohoku earthquake. The authors do not intend the 4, 2000 M 8.0 Sumatra earthquakes. From the relatively large
descriptions to be misinterpreted to indicate that such area mapped in Fig. 1 and sketched in Fig. 2, and from later
responses are inconsequential, all damages can be prevented, maps updated every 6 months, no one can determine exactly
or that all critical vulnerabilities can be identified in advance where the possible earthquake may occur within the circle,
of such large and extreme events. However, we hope the which day it may occur, or the exact magnitude within the
importance, logic, and sequence for implementing actions in specified range (i.e., is it a M 8.0, M 8.3, M 9.0 or other?). As a
response to the prediction can be understood by reviewing result, it is clearly recognized that this alarm is not a red alert
the scenario provided below for the sake of learning from this or indicating an immediate threat of an impending earth-
disaster in order to reduce future disasters. quake for particular communities, but also that the risk is
A possible use of the 2001 prediction is as follows. In mid- clearly larger than the long term risk for which disaster
2001 disaster managers, policy makers, critical infrastructure preparedness actions are normally undertaken, at least for
operators, and other appropriate people and organizations parts of Japan. It is understood that the earthquake could
learn of the M8–MSc prediction identifying a MZ8.0 earth- happen at any time after the alarm is released. As a result, the
quake may occur sometime within the next 5 years some- determination on how to take action in response to this
where within the region identified in Figs. 1 and 2. The alarm cannot be delayed.
probability for this prediction is described as having at least a A review is performed for: (a) the known seismic faults
70% chance of occurring within the large M8 circle of within the alarm region (large circle defined by M8), and
investigation shown in Fig. 2 (f¼30%) and approximately (b) the magnitude of earthquakes they are expected to
22 C. Davis et al. / International Journal of Disaster Risk Reduction 1 (2012) 17–32

generate. Considerations for the potential faults having risk areas participate and to increase the frequency (e.g.,
the capability of generating great earthquakes reveal that semi-annually) during the alarm duration. Emergency
if the earthquake is to occur it will most likely be response plans should be developed or updated at differ-
generated in a zone along the plate boundary contacts ent levels from individuals, families, businesses, local
(Pacific, Eurasian, and Philippine sea; see for example government, prefecture government, national govern-
[44,2]). The narrowed rectangular region shown in Fig. 2 ment, non-government organizations, to even interna-
indicates a 50% chance of the earthquake being generated tional support. Fig. 3 shows that once it is determined
in the vicinity of the contact between the Pacific and the that action is to be taken many of the above listed types of
Eurasian plates, just west of the Japan Trench. This leaves actions should be identified and rapidly reviewed to
a 20% chance of the earthquake being elsewhere in the ensure their practical implementation is justified (follow
prediction region, and may be along one of the other plate the right side of the flow diagram). When justified, they
contact zones, and a 30% chance of the predicted earth- should be implemented in a prudent and efficient
quake not occurring at all. manner.
The Japanese make long-term earthquake forecasts Other types of actions may require more detailed
[24]. The forecasted interplate magnitudes used for seis- analyses to ensure they are prudent to implement. Hazard
mic hazard analyses within the prediction region in assessments for potential M 8.0–M 9.0 or larger earth-
northeast and northwest Japan are less than magnitude quakes are then performed, with a priority and focus
8.0, except for some very low probable earthquakes of up along the Japan Trench. Since time is a critical factor,
to magnitude 8.2 along the Japan Trench. The larger and some previously known hazard assessments such as the
more probable forecasted Nankai, Tonankai, and Tokai tsunami evaluations by Minoura et al. [38] and later by
earthquakes reach up to magnitude 8.5 along southeast Satake et al. [51] are used and even initially applied to
Honshu [52,24], but the annualized M8 predicted prob- other areas while more detailed assessments are per-
ability of occurrence is slightly higher by factors of formed. Results of hazard assessments are distributed to
1.4–2.4. This means that the predicted risks exceed the all potentially impacted communities for their use.
current forecasted risks used for making earthquake The potential impacted communities are recommended
preparedness and the forecasts are to be reconsidered in to have all buildings, critical infrastructure, hazard mitigation
a rapid manner especially in the refined MSc region. and preparedness plans, emergency response and recovery
Evidence for larger earthquakes in northeast Japan is plans, social, economic, housing, commercial buildings, ports,
readily confirmed using the information described in financial systems, insurance coverage, policies, and other
Appendix B, including M 9.0 earthquakes. The prediction related important aspects reviewed to determine gaps
in Fig. 2 then fosters a wider understanding that M 8.2–M between the existing hazard preparedness and the predicted
9.0 earthquakes are considered possible over a much hazard. The potential consequences for each gap are then
wider region than previously used in the forecast models assessed and possible measures that will reduce the potential
and are consistent with historic subduction zone earth- hazard impacts are identified. Table 2 presents example
quakes in other parts of the world. measures that are identified. The identified measures are
This initial review is followed by more detailed hazard assessed for cost effectiveness as shown in Table 2, screened
assessments, which are not described in this scenario, and for the time to implement in relation to alarm duration (TIP),
results in the preliminary finding that there is a high and prioritized for implementation for this alarm. The higher
opportunity for a M 8.0–M 9.0 or larger earthquakes in priories are given to those actions that can be accomplished
northeast Japan. This newly recognized hazard exceeds in the shortest time frame while having the highest prob-
the hazard used for regional infrastructure design and abilities for reducing the greatest potential impacts. In
disaster planning. The previously forecasted earthquake addition, some critical actions that may take longer to
size in south east Japan is consistent with that identified implement than the current alarm are prioritized for imple-
in this prediction. The conclusion: initiate plans for taking mentation with recognition that: (1) the alarm may be
action in response to this alarm. The action planning extended longer than the initial 5-year period; and (2) the
proceeds in parallel with other hazard assessment activ- improvements provide important risk reduction in relation to
ities and each is refined and improved as more informa- the normal background long-term seismic threat.
tion and greater understandings are developed. For example, the consequences of a nuclear meltdown
Many logistical tasks become obvious to implement at are huge; therefore screen the nuclear power plants for
this time including updating emergency response plans, fortifications against tsunami and other hazards that may
coordinating disaster response agencies, coordinating impact them. In this review process the existing 5.7 m
information releases with the media, stocking up on high tsunami protection walls [3] at the Fukushima plant
medical and disaster relief supplies, initiating contracts are identified as having a high potential of being over-
for post-earthquake supply and support, reviewing insur- topped by the projected 8-m high tsunami wave [38]. The
ance coverage, updating procedures and supplies, enga- tsunami wave for an M 9.0 or larger earthquake could be
ging the technical community for more detailed hazard much higher and requires more detailed evaluation. The
evaluations, identifying vulnerabilities, and other emer- Fukushima plant is located approximately 8 m above sea
gency preparedness related activities. In addition, it is level by design, as compared to the approximately 35 m
prudent to perform scenario earthquake and tsunami original ground elevation. The next step is to ask the
drills; many Japanese cities perform these regularly but question: what would be the consequences of a tsunami
it is important to ensure that everyone within the high breach at the Fukushima nuclear power plant? Nuclear
C. Davis et al. / International Journal of Disaster Risk Reduction 1 (2012) 17–32 23

Table 2
Possible scenario actions to reduce seismic risks in response to alarm. Actions are described in the text.

Item Action Ca ($1000) P ($1000) Gain, G ($1000)

1  f¼ 20% 1  f¼ 50%

Nuclear power plants


1a Raise tsunami wall 10,000 500,000 90,000 240,000
1b Protective generator housing 1000 500,000 99,000 249,000
1c Raise wall þ protective housing 11,000 500,000 89,000 239,000

Home, office, maintenance, industrial buildings


2 Anchor furniture, cabinets, computers, equipment, etc. 11 101 9.2 39.5
3a Relocating out of tsunami inundation area, or 500 1,500  200 250
3b Retrofitting structure for tsunami 280 1,500 20 470

Lifeline systems
4 Railway bridge and track 400 700  260  50
5 Water pipe replacement 420 2100 0 630
6 Highway tunnel landslide repair 2000 2000  1600  1000
7 Power transformers 1500 10,000 500 3500
8 Roadway bridge 500 1500  200 250
9 Liquid Fuel Tank 30 2000 370 970

Cultural
10 Nikko temples 50 100  30 0

Disregard/Unaware of Prediction
11 Do nothing 0  518,401  103,680  259,200

power experts develop many scenarios, one which actual losses may exceed $100 billion; [49]). These are
includes the loss of emergency generators. If this scenario clearly cost effective measures to implement and both
was postulated in advance of the Tohoku earthquake alternatives are recommended to initiate immediately with
some options to mitigate, as shown in Table 2 items 1a further evaluation to determine the wall height needed to
and 1b, would include the alternatives: (a) raising the protect against a tsunami resulting from an M 9.0 or greater
tsunami walls and (b) developing protective housing for earthquake. Raising the tsunami wall has a further advan-
generators to resist the potential flood. These alternatives tage of protecting the entire plant. This simple analysis
are then assessed using the optimizations presented by shows that several hundred million dollars on fortification
Molchan [39,40] and further developed and explained by improvements could be cost effectively spent for this alarm
Davis et al. [6,7] using (several hundred billion in improvements would be cost
effective if actual P values were used in the analysis). As a
G ¼ Pð1f ÞC a ð1Þ
result, similar actions and related evaluations are ordered at
where G is the gain, P is the cost of damage prevented by all other nuclear power plants, giving the highest priority to
implementing the action, Ca is the cost of implementing those on the northeast coast, next highest priority in middle,
the action, and f is the probability of false alarm. The southeast, and west coast of Japan, and third highest priority
value of (1 f) can be taken as 0.50 for hazards resulting to all other nuclear power plants in Japan based on their re-
from earthquakes generated along the Japan Trench and evaluated tsunami hazards. Table 2 shows how the second
0.20 for earthquakes generated elsewhere in the predic- priority regions having 1f¼0.20 is cost effective. The third
tion region (i.e., 1  f can be taken as 0.50 for the 8 m or priority is a logical outcome of the tsunami hazard reassess-
higher tsunami wave at the Fukushima nuclear power ment, which was fostered by the 2001 prediction, and is
plant site). A positive G identifies the benefits are greater implemented in addition to rather than as a direct response
than the costs from an economic perspective and the to the prediction.
action is warranted. A negative G indicates a loss and the Another example presented as item 2 in Table 2 is
action may not be warranted. Eq. (1) does not account for provided for anchoring cabinets, computers, and equip-
the likelihood of damage from a successful prediction in ment for a company located in Miyagi Prefecture. The
relation to the larger prediction region relative to the company has adequately seismically strengthened build-
potential source size. Davis [5] describes how to apply ings at an elevation higher than 50 m; tsunami inundation
economic parameters to Eq. (1). is not a concern, but strong shaking can potentially
As shown in Table 2 items 1a–1c, assuming a cost Ca of damage their valuable equipment and the information
$10,000,000 for the tsunami wall raising and $1,000,000 for stored within it. The company can invest Ca ¼$11,000 to
building the protective generator room, and a $500,000,000 anchor their computers, storage cabinets, and other
cost P of damage prevented if the generators are adequately equipment so that it will not topple over during ground
protected to prevent a meltdown then the gain G attained shaking to protect $26,000 in equipment replacement
by taking both actions is $239,000,000 (note P is a very low value and the $75,000 worth of information stored;
value but considered reasonable for a 2001 estimate, the P ¼$101,000. Using Eq. (1) with 1 f¼0.50, the gain is
24 C. Davis et al. / International Journal of Disaster Risk Reduction 1 (2012) 17–32

estimated to be $39,500 and the company decides to and historic value is greater and other similar artifacts are
rapidly anchor the equipment. A similar company having also mitigated elsewhere in Japan.
comparable costs in Shizuoka Prefecture has a gain of In this scenario many other disaster preparedness
$9200 and performs the same actions. actions are undertaken. Similar infrastructure analyses
Table 2 presents several examples for lifeline systems. are performed for bridges, port facilities, water and
Item 4 in Table 2 presents an example railroad located near wastewater treatment plants, commercial and office
the coast in a potential tsunami inundation zone and on buildings, other power plants, homes, etc. considering
liquefiable soils in Iwate Prefecture. A bridge and 1 km of all the earthquake related hazards previously described.
track are identified as being vulnerable to liquefaction and/ The assessments are performed by all types of organiza-
or tsunami damage. There is a parallel line outside the tions from individual owners, municipalities, water and
tsunami and liquefaction hazard zones, but it is slower and sewer systems, all types of transport systems, financial
for some riders may require an additional 2 h/d to use this and insurance institutions, all levels of government, and
route for getting to their destinations. The cost to improve so on. The review is performed throughout the prediction
the soils to prevent liquefaction damage to the tracks and area using appropriate values of (1 f). Based on assess-
bridge is estimated to be Ca ¼$400,000; this mitigation is ment results, the needed actions are prioritized and
also expected to strengthen the bridge to resist tsunami implemented to mitigate hazard vulnerabilities and make
damage. The damage prevented including costs of customer necessary preparations in a manner previously described.
disruption is estimated to be P¼$700,000. The gain is In July 2006, as a result of monitoring seismicity, the
estimated to be $50,000, a net loss, and the company alarm is extended for another five years and the alarm
decides not to perform any mitigation measures, but instead region shifts to the north as shown in Figs. 1 and 2
increases insurance coverage and develops action plans in (technically the alarms in the original circles of investiga-
case this section of line is damaged and also to make rapid tion expire and new alarms are called in circles of
repairs following the event in an attempt to lessen P. investigation to the north). The refined prediction area
Items 5 through 9 in Table 2 present examples for other made by the MSc algorithm remains nearly the same as
lifeline systems. The water pipe in item 5 is critical to that identified in 2001. Confidence in the alarm is
providing supply to hospitals in an emergency and crosses renewed and sustained based on continued successful
through a liquefiable zone. The costs Ca are for replacing the worldwide predictions-in-advance using the M8–MSc
pipe with ductile iron pipe having special earthquake resis- algorithms including the recent May 3, 2006 M 8.1 Tonga
tant joints that allow the pipe to deform under liquefaction earthquake, and several months to years later the April 1,
induced permanent ground deformations without leaking. 2007 M 8.1 Solomon Islands, September 12, 2007 M 8.5
The existing power transformers in item 7 are on soft soil and 8.1 Sumatra, September 29, 2009 M 8.5 Samoa, and
sites that will be subjected to differential settlements and February 27, 2010 M 8.8 Chili earthquakes. As a result, the
render a large portion of the system unusable; their vulner- hazard evaluation process is reviewed once more and
abilities can be mitigated by adding concrete piles as part of identifies that potential large earthquakes in southeast
their foundations. The liquid fuel tanks in item 9 are at a port Honshu are no longer in the prediction region (namely
and unanchored; a tsunami wave may displace and damage Nankai), which somewhat lowers the risk in that region
the tanks making them unusable, but can be mitigated by returning it to the background long-term forecast levels.
adding an anchorage system. Table 2 shows items 5, 7, and 9 Other regions in Hokkaido have increased risks and
to be cost effective in any location within the prediction actions similar to that previously described are imple-
region. The case in item 6 where a landslide can potentially mented on that island. This took place only a few years
block a highway tunnel is not cost effective for this prediction after the 2003 M 8.3 Tokachi-Oki earthquake and tsunami
at any location, but some may still consider its viability based (not detected by the M8 algorithm) struck that island.
on the life safety concerns. The roadway bridge in item 8 is Even though this region was not detected as having a
located along the coast and requires strengthening to resist great earthquake some damage was averted due to the
tsunami forces; and is found to be cost-effective along the preparations in response to the 2001 alarm because
north-east coast of Japan, but not in other areas of the hazard evaluations identified portions of Hokkaido to
prediction region. As a result, bridges under this condition have the potential for shaking and tsunami damage if an
are prioritized for mitigation improvements along the north- earthquake did occur in the northern end of the predic-
east coast. tion region. Preparations include strengthening vulner-
Eq. (1) is also used to assess objects of historical value. The able structures, improving post-earthquake fire fighting
Nikko Temples (located in the Tochigi Province about 50 km capabilities, and adding tsunami barriers. In the Tokachi-
north of Tokyo) are inscribed as one of UNESCOs world Oki earthquake shown in Fig. 1 and Table 1 there were
heritage sites and are evaluated for and determined to need nearly 600 injuries suffered and many homes and other
about $50,000 worth of improvements to protect them buildings were lost.
against about $100,000 in damage, based on Modified In January 2011 the alarm is called off and classified as
Mercalli Intensity (MMI) IV shaking from an earthquake a false alarm. There is no significant reversion of any
along the Japan Trench. Table 2 shows that item 10 is cost- actions taken in response to this alarm, nearly all mea-
effective in this area of the prediction region. However, sures remain in place. On March 11, 2011 a mega-thrust
similar cultural artifacts elsewhere in Japan may not be earthquake strikes northeast Japan and followed by a
cost-effectively mitigated for this prediction based entirely great tsunami. This is the Tohoku earthquake. Due to
on monetary reasons. However, some argue that the cultural the extensive measures undertaken in response to the
C. Davis et al. / International Journal of Disaster Risk Reduction 1 (2012) 17–32 25

alarm shown in Fig. 1, a significant amount of damage five-year duration. As a result, instead of preventing inun-
was averted and numerous lives saved. dation other alternatives should also be addressed for this
alarm such as post-earthquake evacuations, strengthening
3.2. Additional aspects on the scenario use of prediction buildings, moving materials and supplies to higher eleva-
tions, etc. Item 3 in Table 2 considers options for: (a)
The information presented in the previous subsection relocating a structure out of a tsunami inundation area or
clearly identifies how some very high risks could have (b) strengthening an existing concrete building to resist the
been significantly reduced, or in some cases averted tsunami and maintaining important artifacts on floors
altogether, if proper assessments were made using the higher than the expected flooding. For the cases presented
information provided by the alarm. This is irrespective of it is more cost effective to mitigate than relocate. Now
the alarm being called off only weeks in advance of the considering that tsunami barriers may provide cost-effective
Tohoku earthquake. Knowing there was a 30% probability and sufficient protection to communities, the initiation of
for the predicted earthquake not to transpire, even if the such barriers may be warranted based on the alarm in areas
earthquake did not occur by July 2011 the implementa- needing smaller and much easier to build fortifications than
tion of reasonable and prudent actions would have still that in Kamaishi Bay. These could possibly be completed in
helped Japan prepare for the inevitable earthquake strike. time if the projects were given sufficient priority, especially
Application of the process shown in Fig. 3 for the other since the current alarm was extended for five years. How-
past earthquakes predicted by the M8–MSc algorithms ever, the cost-effectiveness should be weighed against other
and shown in Fig. 1 may have also resulted in earthquake alternatives similar to items 3a and 3b in Table 2. Even if
risk reductions in advance of the impending earthquakes. alarm time extensions were not made, the barriers would
For example, the 1986 prediction could have been used by provide protection from great earthquakes expected to
Guam to improve the stability of vulnerable school and occur in the longer-term if that construction process were
hotel buildings that were damaged [8] by the impending sufficient to maintain stability against the Tohoku earth-
August 8, 1993 earthquake. The 1992 prediction could quake and tsunami. This is not to argue if such barriers are
have been used to mitigate vulnerable bridges and hospi- or are not warranted when considering the long-term
tals [42] that were damaged by the October 4, 1994 hazards; further review of this protective strategy is beyond
earthquake and destructive tsunami on Shikotan Island. the scope of this paper. Other large and lengthy projects
The costs in Table 2 use average values and are not based could undergo similar evaluation processes.
on detailed estimates, but instead are considered as reason- A catalog of all possible missed opportunities is not
able order of magnitude estimates based on similar condi- provided herein; only some of the missed opportunities
tions found in developed and urbanized areas around the are exemplified in order to illustrate how the alarm could
world using direct and indirect costs. The estimates could be have been used for improving and accelerating seismic
improved using exact conditions at specific sites in Japan risk reduction during the TIP. In addition, the logistical
with the cost factors for the years (2001 and forward) the actions identified in the previous subsection are nearly
estimates would have been made. Each situation requires its always found to be cost-effective as noted by Davis et al.
own evaluation and may result in different conclusions than [7]. As a result these types of actions were not repeated in
those described in the scenario. However, higher level Table 2.
estimates based on site specific detailed evaluations would It is seen that not all important and critical actions can be
not change the understanding of how to use Eq. (1) for implemented, even for this alarm that went on for nearly 10
decision making purposes supported by the examples given years. This emphasizes the need for accurate seismic hazard
above. Additionally, improvements in decision making can be assessments and continuously implementing risk reduction
obtained with the use of distribution functions on expected measures. The most beneficial use of M8–MSc predictions is
costs rather than averages. to enhance the long-term disaster risk reduction and emer-
The above examples show how the process in Fig. 3 helps gency preparedness measures that should already be on-
to rationalize and prioritize the development of protective going. The scenario illustrates how intermediate-term earth-
actions. However, not all safety measures are justifiable quake predictions can be used to enhance normal hazard
either due to potential losses as exemplified above, lack of mitigation procedures.
resources to implement, or lack of time to implement in The above description is a scenario identifying how
relation to the alarm duration. The lack of resources and time damages may have been prevented if the appropriate
to implement can be exemplified by reviewing tsunami actions were taken in response to the prediction. Unfor-
barriers for protecting entire coast lines such as the world’s tunately no actions were taken in response to the alarm.
largest at Kamaishi Bay. The Kamaishi Bay protective barrier Table 2 item 11 also shows the option of doing nothing
raises 8 m above the sea and was completed in 2008 at a cost during the alarm. The estimates in Table 2 item 11 only
of $1.4 billion after 30 years of work [3]. The hazard analysis include those examples presented in the scenario and
previously described identifies this barrier to have insuffi- listed as items 1 through 10. There is no initial cost for
cient height to protect the city from a tsunami that may doing nothing. However, there is also no investment into
result from a successful prediction. The cost and time to raise improving damage prevention and reducing negative
this wall to approximately double its capacity is too great to impacts from the predicted earthquake. As a result,
accomplish for the initial five-year duration of this alarm. It Table 2 shows a significant loss for all f if no action is
is also inconceivable to consider developing similar walls taken. The harsh reality of this situation was exposed in
elsewhere in response to this alarm considering only a the aftermath of the great Tohoku Japan earthquake.
26 C. Davis et al. / International Journal of Disaster Risk Reduction 1 (2012) 17–32

4. Tohoku earthquake how the earthquake resulted in a large area of the


Japanese island of Honshu being shaken with MMI of VII
On March 11, 2011, only 70 days after the M8 alarm or more, directly affecting nearly 41 million people resid-
for M8.0þ earthquakes was canceled, the M 9.0 Tohoku ing in about 128 different cities. The shaking lasted up to
earthquake struck off the east coast of Honshu, Japan, as 6 min in some locations and generated the largest ever
shown in Figs. 1 and 4, at 2:46 p.m. local time. This was recorded accelerations of 2.7g measured directly in the
one of the five most powerful earthquakes ever recorded. Miyagi prefecture, 75 km from the epicenter.
The earthquake resulted from thrust faulting on or near The seafloor uplift generated a large tsunami wave at sea
the subduction zone plate boundary between the Pacific on the order of 15–20 m high [3] that propagated throughout
and the Eurasian plates [2]. The hypocenter was located at the Pacific Ocean. The tsunami affected more than 2100 km
a depth of about 32 km approximately 129 km east of of the eastern Japan coastline [19] and traveled up to 10 km
Sendai, Honshu, Japan [54]. The fault slipped upwards in inland [50]. Fig. 5 shows the tsunami wave height distribu-
the order of 30–40 m over an area approximately 300 km tion along the coast as measured at official recording stations.
long (along-strike) by 150 km wide (in the down-dip The preliminary reported inundation depth reached up to
direction), lifting up the seafloor as much as 9 m [55]. about 27 m and the run-up height reached up to about 38 m
The main shock was preceded by a M 7.2 foreshock on [20]. The world’s largest tsunami protection wall at Kamaishi
March 9, 2011 and followed by at least six major (M Bay rises 8 m above the water and was overtopped by a 14 m
7.0 þ) and 120 strong (M 6.0 þ) aftershocks. wave [3,23].
The scale of the Tohoku earthquake disaster exceeds The tsunami was far more deadly and destructive than
that of all previous disasters. Fig. 4 and Table 3 identifies the actual earthquake. Entire communities were devastated

Fig. 4. Tohoku earthquake hazard maps. Left: map of seismic shaking intensities (MMI) estimated by the USGS [54]. Right: map of observed tsunami
wave height [22] at official recording stations.

Table 3
Estimated population exposure. Prepared using data obtained from USGS [54].

MMI Total

VII VIII IX X

Damage potential Moderate Moderate/Heavy Heavy Very Heavy


No. of cities 114 (incl. Tokyo) 13 1 (Furukawa) 0 128
Population 34,736,000 5,841,000 255,000 0 40,832,000
C. Davis et al. / International Journal of Disaster Risk Reduction 1 (2012) 17–32 27

and some towns were nearly completely destroyed by the power plant, for purposes of explaining the application of
tsunami. In total, there are more than 19,000 dead or missing, Eq. (1). It is difficult for researchers to acquire post-event
mostly elderly people, and about 5900 reported injuries. data of sufficient detail shortly after an event that is
There are 1,090,000 buildings damaged or destroyed, of useful for the cost analysis presented herein. However,
which 127,197 totally and 232,083 partially-collapsed, item 9 in Table 2 represents an actual case of a liquid fuel
mostly residential, from shaking, tsunami, and fire [43]. Over system performance based on specific observations in
440,000 people have been displaced from their homes by the Ishinomaki, but lack actual cost data; the costs are rough
earthquake, tsunami, fire, and nuclear radiation alert and estimates. Further information for this case is presented
radioactive contamination of the environment [50]. On here to make a tangible link between potential mitigation
March 11, 2011 around 4.4 million households in north- options and actual performance and to show how post-
eastern Japan were left without electricity and 1.5 million event observations can be used to improve the use of
without water. One dam failed, the earth-fill Fujinuma Dam, predictions.
in Fukushima Prefecture killing eight people and causing Figs. 5 and 6 show damage to two liquid fuel storage
other property damages [21]. Numerous fires ignited as a tanks in Ishinomaki City following the tsunami. Fig. 5
result of damage from shaking and the tsunami. In Kesen- shows the foundations for two tanks, one tank is missing
numa oil leakage resulted in a fire that burned several square and the other has slid off of its foundation. Fig. 6 shows
kilometers. An explosion at the Cosmo Oil refinery caused a the missing tank from Fig. 5 that floated away in the
fire that burned for two days [50]. The earthquake and tsunami and was placed on its side on a city street. These
tsunami created an estimated 25 million tons of rubble and tanks were located near the port and could have been
debris in Japan [56]. used for storage of critically needed fuel supplies follow-
All affected nuclear power plants reportedly shut down ing the earthquake and tsunami, if they had not been
safely after the shaking. The Onagawa Nuclear Power Plant damaged. Damages to these tanks and their connecting
had a fire ignite in the turbine building, but did not affect the
reactors. The Tōkai Nuclear Power Plant was shut down and
able to maintain cooling for the reactors. The Fukushima
Nuclear Power Plants I and II were impacted by tsunami
waves that overtopped seawalls and destroyed diesel backup
power systems, leading to severe problems at Fukushima I,
including at least three large explosions and radioactive
leakage. Over 200,000 people were evacuated [56,9]. Damage
to the Fukushima Nuclear Power Plants has resulted in the
second worst nuclear accident in history creating loss of
usable lands, impacts on food and water, and many other
impacts including radiation concerns across the Pacific
Ocean. Losses from damage to the Fukushima nuclear power
plants and resulting radiation may exceed $100 billion [49].
This great earthquake and tsunami has also severely
impacted the local and global economy. Financial institu-
tions estimate the disaster could cost from $122 billion to
$235 billion, or 2.5 to 4 percent of Japan’s gross domestic
product [46]. The government has estimated infrastruc- Fig. 5. Damage to liquid fuel storage tank in Ishinomaki City; tanks
ture damage from the earthquake and tsunami at $190– moved off of foundations.
295 billion. These estimates could make this the world’s
costliest natural disaster [47]. The total insured loss is
between $21 billion on the low end and $34–$42 billion
on the high end, ranking the Tohoku earthquake among
the largest insurance industry losses [50,48]. Impacts
from this earthquake have the potential to permeate
through the global economy for several years to come.
Some of this damage and destruction could have been
reduced, or even eliminated altogether, if appropriate con-
nections were made with disaster managers following the
alarm announcement. If such connections between predictors
and preparers were made, decisions on what and how to take
action could have proceeded in a manner somewhat as
described in Section 3.1 resulting in saved lives and property.

5. Example case: liquid fuel storage tanks


Fig. 6. Damage to liquid fuel storage tank in Ishinomaki City; tank
Section 3.1 presented general actions not specific to floated (photograph provided courtesy of the Japan Water Works
system or site locations, except for the Fukushima nuclear Association).
28 C. Davis et al. / International Journal of Disaster Risk Reduction 1 (2012) 17–32

pipelines mostly resulted due to the lack of anchorage to an algorithm. This is exemplified in the following com-
their foundations. Table 2 shows the costs of damages parison of preparedness actions typically considered for
for direct and business interruption losses is estimated to critical infrastructure and the general public.
be $2,000,000. The estimated cost for taking action in Algorithms having more false alarms may be used to
advance of the disaster in a manner similar to that develop response actions for critical facilities (e.g., dams,
explained for the scenario use of earthquake is estimated nuclear power plants, emergency operations centers, etc.).
to be $30,000. Thus, these estimates indicate the anchor- Examples from the Tohoku earthquake described herein
ing mitigation is relatively inexpensive compared to the show how cost effective measures could have been taken
total estimated losses. The prediction could have been and duplicated during several alarms for critical infra-
used to help identify this needed lifeline improvement. structure. As a result, organizations that operate critical
infrastructure are more likely to tolerate a higher level of
6. Prediction strategies and disaster preparedness false alarms to ensure protective measures are in place
when an actual event occurs. On the other hand, the
The advanced prediction combined with documented general public rapidly becomes insensitive to frequent
results of the great Tohoku earthquake and tsunami disaster alarms, which results in a need to implement strategies
provide a good example for illustrating the possibility of that consider the consequences of alarms having lower f,
implementing different earthquake prediction strategies for but possibly greater n. Actions usually undertaken by the
critical infrastructure and the general public. Examples of general public, except for logistical actions, in places like
critical infrastructure damages resulting from the Tohoku Japan where well developed long term earthquake pre-
earthquake were presented in the previous section. Of the paredness programs exist typically require a much lower f
many different types of critical infrastructure damages two to be cost effective as compared to critical infrastructure;
stand out above the rest as a result of significant secondary this is shown in some of the examples presented herein
damages and cascading costs caused by their failures; they and in Davis et al. [7]. This indicates a strategy applied to
are the Fukushima Nuclear Power Plant and Fujinuma Dam. the general public could include well developed long term
Damages and costs go well beyond the direct impacts from disaster preparedness programs, periodic enhancement of
their failures and impacts will linger for years, if not decades, logistical actions, and earthquake prediction alarms hav-
to come. As previously identified, losses from the Fukushima ing lower f as compared to alarms applied to critical
Nuclear Power Plant may exceed $100 billion. Total loss infrastructure. As a result, the error tradeoffs that inher-
estimates from the Fujinuma Dam are not available, but may ently exist in an earthquake prediction algorithm can be
reach hundreds of millions of dollars based on other similar calibrated to provide multiple algorithm versions for
types of problems (including property loss, casualties, litiga- more efficient use with critical infrastructure and the
tion, loss of irrigation, business impact, replacement costs, general public. This has not yet been done but is the focus
etc.). These losses exceed the costs to prevent such damage of further study. The strategies described herein identify
by orders of magnitude; see Table 2 for the nuclear example. how earthquake predictions can be incorporated along
In such cases, it is cost effective to do everything possible to with other risk reduction methodologies to enhance
ensure these types of failures do not occur. Using Eq. (1) to disaster preparedness.
estimate f at the breakeven point when G¼0 identifies that it At the same time disaster managers and policy makers
was cost effective to take action for the Fukushima nuclear have certain freedoms to choose appropriate tradeoffs on
power plant with a 99.99% probability of false alarm. The how to utilize the alarm information for making decisions
breakeven point for the dam failure will not reach the same and selecting the appropriate set of temporal and perma-
levels, but could easily result at f¼80–90%; dams located nent preparedness measures to undertake. The alarms
above more populated areas may reach a break even at described herein provide examples on how such tradeoffs
f¼99%. As a result, critical infrastructure should: (1) review and decision making could have been utilized if the
the earthquake hazards on a continual basis and implement proper links between earthquake predictors and disaster
necessary mitigations to increase the probability that failures managers were made. As previously described, the con-
do not occur; (2) continually maintain normal emergency centration of sources changed marginally, and only by
preparedness measures, and (3) escalate emergency prepa- 1.5%, but the ‘‘black box’’ code for the M8 algorithm, fixed
redness and response activities when expected risks exceed in 1992, cut off the alarm for M8.0þ earthquakes purely
background levels, including changing priorities and applica- for scientific purposes. If the alarm were being used for
tion of temporary measures. Earthquake prediction strategies actual disaster preparation purposes, a reasonable person
can be implemented to aid in the above mentioned disaster could have made a decision to continue with the M8.0þ
preparedness activities needed for critical infrastructure. alarm, at least for certain critical infrastructure and
Developers of earthquake prediction algorithms have sectors within the regions of concern. At the same time,
certain freedom to choose appropriate tradeoffs between a concurrent alarm for predicting M8.5þ and M9.0þ
different kinds of prediction errors to allow for adequate earthquakes remained in place, but was not part of the
preparedness decisions to be made [40,25,7]. For a given Global testing [34] and the algorithm version, being
magnitude range and TIP, different earthquake prediction within the development stage, was too recent to provide
algorithms can be created to adjust the rate of false useful values of f to use for decision making purposes.
alarms f and failures to predict n. Disaster preparedness Nonetheless, under the circumstances in relation to the
can be more flexible and efficient if earthquake predic- M8.0þ alarm and knowing the very high values of f that
tions were carried out in parallel with several versions of can be tolerated for certain critical infrastructure, a policy
C. Davis et al. / International Journal of Disaster Risk Reduction 1 (2012) 17–32 29

maker could have chosen in this specific case to enact the approximation of prediction (i.e., the confidence in MSc is
M8.5þ and M9.0þ alarms from a limited set of very slightly lower than that for M8). In the timeframe from
critical locations. Unfortunately, these and other related 1985 to 2011 there have been 19 target earthquakes in
decisions were not able to be contemplated due to the the M8 investigation regions and 14 of them have been
lack of communication between earthquake predictors successfully predicted by M8. In the timeframe from 1992
and disaster managers. However, these last minute deci- to 2011 there have been 17 target earthquakes in the
sions would not have significantly changed the final combined M8–MSc investigation regions and 8 of them
outcome, but the outcome and resulting devastation have been successfully predicted by MSc and 12 of them
could have been modified, at least somewhat, if the have been successfully predicted by M8 [34,35]. These
M8.0þ alarm were used to implement cost-effective results indicate the success rate for M8 is at least 71% and
mitigation actions in the years, months, and days prior 47% for MSc (rounded values of 70% and 50% are used in
to March 11, 2011. The authors hope that this case study the main body for calculations in Table 2).
shows the importance of using credible earthquake pre- Molchan and Romashkova [41] performed a detailed
dictions to help reduce future potential damage from the study on the M8 prediction results for magnitudes up to
world’s largest earthquakes and helps to create better 8.5 and 8.7. Their results are slightly different than that
communication links between earthquake predictors, dis- identified by Kossobokov [34,35] because Molchan and
aster managers, and policy makers. Romashkova [41] re-evaluated the entire earthquake catalog,
finding some earthquake magnitudes have been revised (e.g.,
7. Conclusions 1995 Iturup earthquake), and they reviewed prediction
successes based on varying magnitude threshold and predic-
Earthquake predictions covering the area of north- tion space. Molchan and Romashkova [41] report success
eastern Honshu identified a 70% chance of a magnitude rates ranging from 52 to 60% for the M8 algorithm, but did
8.0 or greater earthquake to occur sometime between not review the MSc algorithm. The primary differences
mid-2001 and mid-2011 using the combined M8–MSc between the Molchan and Romashkova [41] and Kossobokov
algorithm. These predictions accurately identified the [34,35] results can be understood as follows. The Molchan
March 11, 2011 M 9.0 Tohoku earthquake and could have and Romashkova [41] results represent an attempt to strictly
been utilized to make important preparations to reduce identify success of the M8 algorithm that is purely defend-
damage from the actual earthquake. Damage prevention able at a scientific level but have huge margins of reserved
could have included the resulting release of radiation uncertainty (their choice of k¼ 65, the ratio of total area for
from the Fukushima nuclear power plant, among many all CI’s to one CI, appears to heavily overestimate the number
other things ranging from personal evacuations to trans- of degrees of freedom due to a formal evidently incorrect
port corridors to better financial systems management split of space into a set of non-intersecting areas). Molchan
(banking, insurance, etc.). and Romashkova [41] also eliminate some of the earthquakes
The primary reasons for not using the prediction for reported as successful predictions by Kossobokov [34,35] and
improving preparations in advance of the Tohoku earth- add some earthquakes that fall inside the magnitude–spatial
quake were postulated to include: (1) inadequate links domain of global testing due to regional bias in magnitude
between emergency managers and the earthquake pre- determination (e.g., for earthquakes from New Ireland and
diction information; and (2) no practicing application of New Britain regions, P.N.G., and the 2008 Wenchuan, China
existing methodologies to guide emergency preparations earthquake). Note that sizing earthquakes remains a very
and policy development on how to make decisions based difficult and rather uncertain problem of contemporary
on information provided for an intermediate-term mid- seismology. Therefore, the success rates as reported by
dle-range earthquake prediction having limited but Kossobokov [34,35] and Molchan and Romashkova [41] are
known accuracy. Processes were identified [39,40,6,7] both of significant importance to disaster managers because
and examples provided for the Tohoku earthquake on earthquakes were identified to meet the prediction criteria or
how reasonable, prudent, and cost effective decisions can within a very close margin that resulted in impacts that may
be made in a rapid manner to reduce damaging earth- have been mitigatable with knowledge of the prediction.
quake effects. Using information provided by the M8–MSc Since Molchan and Romashkova [40] do not identify results
algorithms has the capability of reducing the worldwide for MSc, the results reported by Kossobokov [34,35] are used
impacts from the largest earthquakes. for analysis in this paper. However, since the Molchan and
Romashkova [41] most conservative minimum reported
Appendix A. M8 and MSC algorithms success rate for M8 is about the same as the one for M8–
MSc given by Kossobokov in [34,35] is reflected in Table 2 in
A.1. M8 and MSc algorithms the right hand column.
The M8 and MSc algorithms are summarized below.
Algorithm M8 provides intermediate-term middle- Fig. 1 shows the case study prediction from the combined
range predictions having characteristic alarm durations algorithms encompassing the area where the March 11,
of about 5 years [25]. The algorithm MSc provides a 2011 M 9.0 Tohoku Japan occurred.
second approximation to M8, strongly reducing the alarm
area to a narrow or exact range. Predictions are made first A.2. Algorithm M8
by M8. Then, the areas of alarm are reduced by MSc at the This algorithm was designed by retrospective analysis
cost that some earthquakes are missed in the second of seismicity preceding the greatest (MZ8.0) earthquakes
30 C. Davis et al. / International Journal of Disaster Risk Reduction 1 (2012) 17–32

worldwide, hence its name [26,27,31]. In 1992 an experi- declaration. A ‘‘sufficiently large’’ size of clusters suggests
ment was launched aimed at real-time earthquake pre- large scale correlations in the recent times [33].
diction on a global scale using the M8 algorithm [18]. The Fig. 1 illustrates the MSc algorithm application from
area targeted for prediction in M8.0þ magnitude range 1985 to 2011 in an area stretching from Kamchatka to the
(i.e., from magnitude 8.0 to 8.8) covers 80% of the major Marianas.
seismic belts—the territory for which the earthquake
catalog is sufficiently complete to apply the algorithm.
This territory is scanned by overlapping circles with Appendix B. Past studies recognizing rare events
diameter of 121 1333 km (see Fig. 1). Predictions are
made separately for each circle. Pattern recognition methods were employed to iden-
The algorithm for issuing M8 alarms is illustrated in tify earthquake prone areas for the first time in the 1970s
Fig. 1, showing the case history of several predictions [12,13], and were later used to define areas prone to great
stretching from Kamchatka to the Marianas. Within each earthquakes having MZ8.2 [17]. As shown in Fig. B1,
CI several characteristics of seismicity are monitored. The major seismic belts were divided into D and N segments
number of earthquakes (seismic flux rate) and its devia- where, respectively, earthquakes with MZ8.2 are possi-
tion from long-term trend (differential of rate, i.e. accel- ble (even yet unknown) and not expected to nucleate. The
eration) are complemented with the linear concentration division is based on pattern recognition of rare events,
of sources to depict the expanded phase space of seismic which is not to be mixed with statistical pattern recogni-
activity averaged over a large time window (six years); tion. This methodology was developed by the School of I.
each measure is computed in two magnitude intervals to Gelfand and is successful in a multitude of applications,
make description of seismicity more robust giving six among them is predicting earthquake prone structures in
functions. The fourth characteristic depicting earthquake different regions and magnitude ranges [15].
clustering over one year provides a total of seven func- Fig. B1 identifies the location for the March 11, 2011 M
tions used for scoring the M8 algorithm. An alarm, a TIP of 9.0 Tohoku Japan earthquake. As seen in Fig. B1, this
an earthquake occurring, is issued for five years when earthquake did occur on the D segment as shown by the
clustering and at least five other characteristics (a total of arrow. This example provides another confirmation point
6 out of 7 characteristics) become ‘‘very large’’ at approxi- on the validity of pattern recognition methods in assess-
mately the same time. The algorithm is re-evaluated ment of seismic hazard.
every 6 months and updates to the alarm are made as Paleo-tsunami studies reported in the past decade
necessary. Updates may include calling the alarm off in identified the potential of great and unusual earthquakes
any or all circles of investigation, or calling new alarms for to strike Japan off the north east coast of Honshu in the
other circles of investigation. From a non-technical per- Fukushima, Miyagi, and Iwate Prefectures. Inter-plate
spective these updates look like adjustments to the earthquakes of M 8.1–8.3 were considered highly prob-
alarm regional space and timeframes. A more detailed able [38,51] and magnitudes up to 9 were identified as
and technical description of the M8 algorithm can be possible [4]. Historic tsunami waves were estimated to be
found at the web site http://www.mitp.ru/en/m8pred. on the order of 8 m high [38] for an M 8.3 earthquake.
html and Keilis-Borok and Kossobokov [26,28,29]. Great earthquakes of M 8.0–8.5 have been recognized
with probabilities of 50–90% of occurring in 30 years in

A.3. Algorithm MSc


Algorithm MSc or ‘‘The Mendocino Scenario’’ [37] was
designed by the retroactive analysis of the regional
seismic catalog prior to the Eureka earthquake (1980,
M¼7.2) near Cape Mendocino in California, hence its
name. MSc is applied whenever the seismicity is suffi-
ciently high to allow the algorithm to be used. It is aimed
at the reduction of the territory where a strong earth-
quake is already predicted by some other algorithm.
Seismic activity in the territory of the original alarm is
analyzed in more details, in the overlapping small cells
and smaller magnitude range. The alarm is confined to the
area where seismicity is high but irregular, interrupted by
short intervals of quiescence [33]. Additional technical
details describing applications of the MSc algorithm are
found at the web site http://www.mitp.ru/en/mcspred.
html and Kossobokov et al. [37]
The MSc algorithm prediction is localized to the spatial
projection of all recent ‘‘sufficiently large’’ clusters of Fig. B1. Earthquake prone segments of major seismic belts. D points are
prone to nucleate earthquakes of M Z 8.2, N points are where strong
square cells being in a state of ‘‘anomalous quiescence’’.
earthquakes (MZ 8.2) are considered not possible. The arrow points to
‘‘Anomalous quiescence’’ suggests a locally high level of the March 11, 2011 Tohoku earthquake showing it is located at a D
seismic activity during formation of a TIP and after its point. Modified from Gvishiani et al. [17].
C. Davis et al. / International Journal of Disaster Risk Reduction 1 (2012) 17–32 31

southeast Honshu along the Nankai Trough shown in [15] Gorshkov AI, Kossobokov VG, Soloviev AA. Recognition of earth-
Fig. 2 [52,24]. quake prone areas. In: Keilis-Borok VI, Soloviev AA, editors. Non-
linear Dynamics of the Lithosphere and Earthquake Prediction.
The previously mentioned studies used in combination Heidelberg: Springer; 2003. p. 239–310 (Chapter 6).
with the July 2001 earthquake prediction described [16] Gvishiani AD, Zelevinsky AV, Keilis-Borok VI, Kosobokov VG. Study
herein indicates that the initial prediction, and all semi- of the violent earthquake occurrences in the Pacific Ocean Belt with
the help of recognition algorithms. Proceedings of the Academy of
annual updates through January 2011, identified the
Sciences of the USSR. Physics of the Earth 1978;9:
potential for earthquakes greater than M 8.2 to occur. In 31–42 (in Russian).
the prediction region north of the Sagami Trough (see [17] Gvishiani AD, Zelevinsky AV, Keilis-Borok VI, Kossobokov VG.
Fig. 2) M 8.2 is the maximum magnitude considered to Recognition of the areas where the strongest earthquakes
(M Z8.2) may occur. In: Keilis-Borok VI, Levshin AL, editors.
occur, and their probabilities of occurrence are very low Methods and Algorithms for Interpretation Seismiolgical Data,
[52,24]. Unfortunately, prior to the Tohoku earthquake Computational Seismology, vol. 13. New York: Allerton Press;
the seismic hazards used for implementing protective 1980. p. 28–38.
[18] Healy JH, Kossobokov VG, Dewey JW. A Test to Evaluate the
measures in northeast Honshu were primarily based on Earthquake Prediction Algorithm, M8 United States Geological
magnitudes ranging from 7.4 to 7.7, and associated Survey. Open Filre Report 92-401; 1992, 23pp.
tsunami wave heights of 4–5 m [3]. [19] hubpages.com. The 2011 Magnitude 9.0 Great Tohoku Earthquake
(Northeast Honshu, Japan), /http://hubpages.com/hub/the-2011-
tohoku-earthquake-in-northeast-honshu-japanS [accessed May 1,
References 2011].
[20] IOC/UNESCO. Casualties by the Earthquake and Tsunami. Bulletin
no. 23, as of 2 May 2011. The National Police Agency. /http://www.
[1] Allen CR (Chairman), Edwards W, Hall WJ, Knopoff L, Raleigh CB,
ngdc.noaa.gov/hazard/tsunami/pdf/20110311_iocbulletin.pdfS
Savit CH, et al. Predicting Earthquakes: A Scientific and Technical
[accessed May 2, 2011].
Evaluation—With Implications for Society. Panel on Earthquake
[21] Japan Commission on Large Dams (JCOLD). 4th Quick Report on
Prediction of the Committee on Seismology. Assembly of Mathe- Dams, April 1, 2011.
matical and Physical Sciences, National Research Council. U.S. [22] Japan Meteorological Agency (JMA). The 2011 off the Pacific Coast
National Academy of Sciences. Washington, D.C.; 1976. of Tohoku Earthquake Observed Tsunami. Observed Time and Height.
[2] Bird P. An updated digital model of plate boundaries. Geochemistry /http://www.jma.go.jp/jma/en/2011_Earthquake/2011_Earthqua
Geophysics Geosystems 2003;4(3): ke_Tsunami.pdfS [accessed May 2, 2011].
1027, http://dx.doi.org/10.1029/2001GC000252. [23] Japan Probe. World’s Deepest Breakwater Failed to Stop Tsunami. /
[3] Cyranoski D. Japan faces up to failure of its earthquake prepara- http://www.japanprobe.com/2011/03/14/worlds-deepest-breakwa
tions. Systems for forecasting, early warning and tsunami protec- ter-failed-to-stop-tsunami/S [accessed May 8, 2011].
tion all fell short on 11 March. Nature 2011;471: [24] J-SHIS (Japan Seismic Hazard Information Stations). /http://www.
556–7 (31 March 2011). j-map.bosai.go.jp/j-shis/index_en.htmlS [accessed May 25, 2011].
[4] Daily Yomiuri Online. Was 896 Tohoku Quake Japan’s Strongest? [25] Keilis-Borok VI. Earthquake prediction: state-of-the-art and emer-
The Yomiuri Shimbun; October 12. /http://www.tsunami-alarm- ging possibilities. The Annual Review of Earth and Planetary
system.com/fileadmin/media/press/articel/en/Daily-Yomiur Sciences 2003;30:1–33.
i-12_10_2007-Was-869-Tohoku-quake-Japans-strongest.pdfS [26] Keilis-Borok VI, Kossobokov VG. A complex of long-term precursors
[accessed May 2, 2011]; 2007. for the strongest earthquakes in the world. In: Proceedings 27th
[5] Davis CA. Loss functions for temporal and spatial optimizing of geological congress. Nauka, Moscow; 1984, vol. 61, p. 56–66.
earthquake prediction and disaster preparedness. Pure and Applied [27] Keilis-Borok VI, Kossobokov VG. Periods of high probability of
Geophysics, http://dx.doi.org/10.1007/s00024-012-0502-8, in occurrence of the world’s strongest earthquakes. Computational
press. Seismology, vol. 19. New York: Allerton Press; 1987 (p. 48–57).
[6] Davis C, Keilis-Borok V, Molchan G, Shebalin P, Lahr P, Plumb C. [28] Keilis-Borok VI, Kossobokov VG. Premonitory activation of earth-
Earthquake Prediction and Disaster Preparedness: Interactive Ana- quake flow: algorithm M8. Physics of the Earth and Planetary
lysis. Abstract Presented at the AGU Annual Meeting. San Francisco, Interiors 1990;61:73–83.
CA; December 14, 2004. [29] Keilis-Borok VI, Kossobokov VG. Times of increased probability of
[7] Davis C, Keilis-Borok V, Molchan G, Shebalin P, Lahr P, Plumb C. strong earthquakes (M Z7.5) diagnosed by algorithm M8 in Japan
Earthquake prediction and disaster preparedness: interactive ana- and adjacent territories. Journal of Geophysical Research
lysis. ASCE Natural Hazards Review 2010;11(4):173–84. 1990;95(B8):12413–22.
[8] Earthquake Engineering Research Institute (EERI). The Guam Earth- [30] Keilis-Borok V, Soloviev AA. Variations of trends of indicators
quake of August 8, 1993. EERI Special Earthquake Report; October describing complex systems: change of scaling precursory to
1993, available at /http://www.eeri.org/lfe/pdf/guam_1993_eeri_ extreme events. CHAOS 2010;20:033104.
preliminary_report.pdfS [accessed July 9, 2011]; 1993. [31] Kossobokov VG. The test of algorithm M8. In: Sadovsky MA, editor.
[9] ESRI. 2011 Tōhoku Earthquake and Tsunami. /http://www.esri. Algorithms of Long-Term Earthquake Prediction. Lima, Peru:
com/services/disaster-response/japan-earthquake-tsunami-2011- CERESIS; 1986. p. 42–52.
map/nuclear-impact-map.htmlS [accessed May 2, 2011]. [32] Kossobokov VG. Quantitative earthquake prediction on global and
[10] Eos. UNESCO/IASPEI Working Group on a Code of Practice for regional scales. In: Recent geodynamics, georisk and sustainable
Earthquake Prediction. American Geophysical Union, vol. 65, no. development in the Black Sea to Caspian Sea region. In: Ismail-
7, Feb. 14, 1984. Reprint from International Union of Geodesy and Zadeh A, editor. American institute of physics conference proceed-
Geophysics, Chronicle, 165, p. 115–124; 1984. ings, vol. 825. Melville: New York; 2006. p. 32–50.
[11] Gabrielov AM, Dmitrieva OE, Keilis Borok VI, Kossobokov VG, [33] Kossobokov VG. Quantitative Earthquake Prediction: Twenty Years of
Kouznetsov IV, Levshina TA, et al. Algorithms of Long Term Real-Time Application and Testing, presentation, available at: /www.
Earthquakes’ Prediction. Lima (Peru): CERESIS; 1986 61pp. lmd.ens.fr/E2C2/class/e2c2@Comorova.pptS [accessed April 25, 2007].
[12] Gelfand IM, Guberman ShA, Izvekova ML, Keilis-Borok VI, Ranzman [34] Kossobokov V. Are mega earthquakes predictable? Atmospheric
EYa. On criteria of high seismicity. Transactions of the Academy of and Oceanic Physics 2011;46(8):951–61 (Izvestiya, Pleiades
Sciences of the USSR 1972;202(6):1317–20. Publishing, Ltd., English translation).
[13] Gelfand IM, Guberman ShA, Izvekova ML, Keilis-Borok VI, Ranzman [35] Kossobokov V. Global and Regional Monitoring Aimed at Earth-
EYa. Criteria of high seismicity determined by pattern recognition. quake Forecast/Prediction. Advanced School on Understanding and
Tectnophysics 1972;13(1/4):415–22. Prediction of Earthquakes and Other Extreme Events in Complex
[14] Ghil M, Yiou P, Hallegatte S, Malamud BD, Naveau P, Systems, 26 September–8 October 2011, Trieste: ICTP, 2265-22;
Soloviev A, et al. Extreme events: dynamics, statistics and predic- 2011b. p. 70. Presentation, available at: /http://cdsagenda5.ictp.it/
tion. Nonlinear Processes in Geophysics 2011;18: askArchive.php?base=agenda&categ=a10170&id=a10170s52t61/
295–350 /www.nonlin-processes-geophys.net/18/295/2011/S (pub- lecture_notesS.
lished by Copernicus Publications on behalf of the European Geos- [36] Kossobokov V, Shebalin P. Earthquake prediction. In: Keilis-Borok
ciences Union and the American Geophysical Union). VI, Soloviev AA, editors. Nonlinear Dynamics of the Lithosphere and
32 C. Davis et al. / International Journal of Disaster Risk Reduction 1 (2012) 17–32

Earthquake Prediction. Berlin-Heidelberg: Springer-Verlag; 2003. K0YJ20110321S, March 21, 2011. Reporting by Chizu Nomiyama in
p. 141–207. Tokyo, Editing by Sugita Katyal [accessed May 2, 2011]; 2011.
[37] Kossobokov VG, Keilis-Borok VI, Smith SW. Localization of inter- [47] Reuters. Factbox: Japan Disaster in Figures. /http://www.reuters.
mediate-term earthquake prediction. Journal of Geophysical com/article/2011/04/09/japan-figures-idUSL3E7F82NG20110409S,
Research 1990;95:19763–72. April 9, 2011 [accessed May 2, 2011]; 2011.
[38] Minoura K, Imamura F, Sugawara D, Kono Y, Iwashita T. The 869 [48] Reuters. Special Report: Japan Quake Reveals Cracks in Insurance
jogan tsunami deposit and recurrence interval of large-scale System. /http://news.yahoo.com/s/nm/20110405/ts_nm/us_insur
tsunami on the Pacific Coast of Northeast Japan. Journal of Natural ance_earthquakesS, April 5, 2011, Ben Berkowitz and Myles
Disaster Science 2001;23(2):83–8. Neligan [accessed May 2, 2011]; 2011.
[39] Molchan GM. Earthquake prediction as a decision making problem. [49] Reuters. UPDATE 7-Tepco Chief Quits After $15 bln Loss On Nuclear
Pure and Applied Geophysics 1997;149:233–47. Crisis. /http://www.reuters.com/article/2011/05/20/tepco-idUS
[40] Molchan GM. Earthquake prediction strategies: a theoretical ana- L4E7GK18U20110520S, May 20, 2011, Nathan Layne and Taiga
lysis. In: Keilis-Borok VI, Soloviev AA, editors. Nonlinear Dynamics Uranaka [accessed May 21, 2011]; 2011.
of the Lithosphere and Earthquake Prediction. Heidelberg: [50] RMS. Estimating Insured Losses from the 2011 Tohoku, Japan
Springer; 2003. p. 209–37 (Chapter 5). Earthquake and Tsunami. RMS Special Report, April 2011.
[41] Molchan G, Romashkova L. Earthquake prediction analysis based [51] Satake K, Sawai Y, Shishikura M, Okamura Y, Namegaya Y, Yamaki
on empirical seismic rate: the M8 algorithm. Geophysical Journal
S. Tsunami Source of the Unusual AD 869 Earthquake off Miyagi,
International 2010:
Japan, Inferred from Tsunami Deposits and Numerical Simulation
1–13, http://dx.doi.org/10.1111/j.1365-246X.2010.04810.x.
of Inundation. American Geophysical Union. Fall Meeting 2007,
[42] National Oceanic and Atmospheric Administration (NOAA). Geolo-
abstract #T31G-03; 2007.
gic hazards Photos. Earthquake Events, vol. 2. Available at /ftp://
[52] Sekiguchi K. Regional Initiative: Japan. Presentation. GEM Outreach
ftp.ngdc.noaa.gov/hazards/cdroms/geohazards_v2/document/
647025.htmS [accessed July, 9, 2011]; 1994. Meeting, 8–10 June 2009, Hohenkammer, Germany, available at /
[43] National Police Agency of Japan. Damage Situation and Police http://www.globalquakemodel.org/system/files/Regional%20Initia
Countermeasures associated with 2011 Tohoku district—off the tives%20-%20Japan.ppsS [accessed May 25, 2011]; 2009.
Pacific Ocean Earthquake January 6, 2012 /http://www.npa.go.jp/ [53] Seismological Society of America. Guidelines for earthquake pre-
archive/keibi/biki/higaijokyo_e.pdfS. dictors. Bulletin of the Seismological Society of America
[44] National Research Institute for Earth Science and Disaster Preven- 1983;73(6):1955–6 (December 1983).
tion (NIED). 2004 Mid Niigata Earthquake, /http://www.hinet. [54] United States Geological Survey. Magnitude 9.0—Near The East
bosai.go.jp/topics/niigata041023/index_e.htmlS. Background, Coast of Honshu, Japan. /http://earthquake.usgs.gov/earthquakes/
Implication of 2004 mid Niigata Earthquake, Figure 2 [accessed eqinthenews/2011/usc0001xgp/S [accessed April 30, 2011]; 2011.
May 11, 2011]; 2004. [55] United States Geological Survey. Japan Lashed by Powerful Earth-
[45] Peresan A, Kossobokov V, Romashkova L, Panza GF. Intermediate- quake, Devastating Tsunami. /http://soundwaves.usgs.gov/2011/
term middle-range earthquake predictions in Italy: a review. Earth- 03/S [accessed May 2, 2011]; 2011.
Science Reviews 2005;69(1–2):97–132. [56] Wikipedia. Tōhoku Earthquake and Tsunami. /http://en.wikipedia.
[46] Reuters. Factbox: Japan Disaster in Figures. /http://www.reuters. org/wiki/2011_T%C5%8Dhoku_earthquake_and_tsunami#cite_
com/article/2011/03/21/us-japan-quake-numbers-idUSTRE72 note-86S [accessed May 2, 2011].

Das könnte Ihnen auch gefallen