Transcript PREDICTING EARTHQUAKES & EARTHQUAKE HAZARDS: WHY SO LITTLE SUCCESS? Seth Stein Northwestern University “Only fools and charlatans predict earthquakes” Charles Richter (1900-1985) Tohoku, Japan 2011 M 9.1 QuickTime™ and a decompressor are needed to.

PREDICTING EARTHQUAKES &
EARTHQUAKE HAZARDS: WHY SO LITTLE
SUCCESS?
Seth Stein
Northwestern
University
“Only fools and
charlatans predict
earthquakes”
Charles Richter
(1900-1985)
Tohoku, Japan 2011 M 9.1
QuickTime™ and a
decompressor
are needed to see this picture.
Scientists will “be able to predict
earthquakes in five years.”
Louis Pakiser
U.S. Geological Survey, 1971
1970’s optimism
“We have the technology to develop
a reliable prediction system already
in hand.”
Alan Cranston, U.S. senator, 1973
“The age of earthquake prediction is
upon us”
U.S. Geological Survey, 1975
Similar in Japan,
China, USSR
Meaningful prediction involves specifying the location,
time, & size of an earthquake before it occurs
Long-term forecast
- Use earthquake history to predict next one
- Use rate of motion accumulating across fault and
amount of slip in past earthquakes
Short-term prediction
-Find precursors - changes in earth before earthquakes
consistently resolvable from normal variability
Despite some claims, no reliable method yet…
Need to do consistently better than expected by
chance from known statistics of earthquakes in
an area
Postdictions - Texas sharp shooter
Shoot at barn and then draw target
around bullet holes
SAN FRANCISCO
EARTHQUAKE
April 18, 1906
3000 deaths
28,000 buildings
destroyed
(most by fire)
$10B damage
“The whole street was
undulating as if the waves
of the ocean were coming
toward me.”
“I saw the whole city
enveloped in a pile of dust
caused by falling buildings.”
“Inside of twelve hours half
the heart of the city was
gone”
Motion along ~ 500 km of
previously unrecognized
San Andreas Fault
~ 4 m of ground motion
West side moved north
USGS
ELASTIC REBOUND
Over many years, rocks on opposite sides of the fault move,
but friction on fault "locks" it and prevents slip
Eventually strain accumulated overcomes friction,
and fault slips in earthquake
Took 60 years to figure out why this happens!
EARTH’S OUTER SHELL - PLATES
Plates
move at few
cm/yr
San
Andreas
fault:
boundary
between
Pacific &
North
American
plates
Hard to predict when block will slip
PARKFIELD, CALIFORNIA SEGMENT OF SAN ANDREAS
M 5-6 earthquakes about every 22 years: 1857, 1881,
1901, 1922, 1934, and 1966
In 1985, expected next in 1988; U.S. Geological Survey
predicted 95% confidence by 1993
Occurred in 2004 (16 years late)
Discounting
misfit of
1934 quake
predicted
higher
confidence
Science, 10/8//04
"Parkfield is geophysics' Waterloo.
If the earthquake comes without
warnings of any kind, earthquakes
are unpredictable and science is
defeated. " (The Economist)
No precursors in seismicity
(foreshocks), strainmeters,
magnetometers, GPS, creepmeter
$30 million spent on “Porkfield”
project
GPS: GLOBAL
POSITIONING SYSTEM
Satellites transmit radio
signals
Receivers on ground
record signals and find
their position
from the time the
signals arrive
Find mm/yr motions
from changes in
position over time
Stein & Wysession, 2003
San Andreas: GPS site motions show deformation
accumulating that will be released in future
earthquakes
Like a
deformed
fence
GPS SLIP RATE
35 mm/yr
Z.-K. Shen
Over time, slip in
large earthquakes
adds up to plate
motion
About 35 mm/yr
motion between
Pacific and North
America shown by
offset stream
Expect large
earthquakes about
every
4 m / (35 mm/yr) or
115 years
Last one here in
1857…
San
Andreas
Fault
“We are predicting another massive
earthquake certainly within the next 30
years and most likely in the next decade
or so.” W. Pecora, U.S. Geological
Survey Director, 1969
1975 PALMDALE BULGE – uplift reported
SAF
USGS director stated that “a great earthquake” would occur “in
the area ... possibly within the next decade” that might cause up
to 12,000 deaths, 48,000 serious injuries, 40,000 damaged
buildings, and up to $25 billion in damage. California Seismic
Safety Commission stated that “the uplift should be considered
a possible threat to public safety” and urged immediate
preparations…
35 years later, nothing yet..
WHY CAN’T WE PREDICT EARTHQUAKES?
So far, no clear evidence for consistent precursors before
earthquakes.
Maybe lots of tiny earthquakes happen frequently, but only a few grow
by random process to large earthquakes
In chaos theory, small perturbations can have unpredictable large
effects - flap of a butterfly's wings in Brazil might set off a tornado in
Texas
If there’s nothing special about the tiny earthquakes that happen to
grow into large ones, the time between large earthquakes is highly
variable and nothing observable should occur before them.
If so, earthquake prediction is either impossible or nearly so.
At present
No reliable method of predicting
earthquakes
No present approaches seem promising
Barring conceptual breakthrough,
earthquake prediction appears unlikely
soon
“It is hard to predict earthquakes,
especially before they happen.”
Hiroo Kanamori
To design buildings and other mitigation
measures, make maps that try to predict the
hazard defined as maximum shaking
(acceleration) that will occur in some time period
Frankel et al., 1996
Map shows central US as hazardous as California
How credible is it?
Underprediction
2001 hazard map
2010 M7 earthquake shaking
much greater than predicted
for next 500 years
http://www.oas.org/cdmp/document/seismap/haiti_dr.htm
6 mm/yr fault motion
Even at fast moving
(80 mm/yr ) &
seismically very active
plate boundaries with
long seismic history,
hard to assess
earthquake hazard
Map assumed high
hazard in Tokai “gap”
2011 M 9.1 Tohoku,
1995 Kobe M 7.3 &
others in areas
mapped as low hazard
Geller
2011
Planning assumed maximum magnitude 8
Seawalls 5-10 m high
Tsunami runup
approximately twice
fault slip (Plafker, Okal
& Synolakis 2004)
M9 generates much
larger tsunami
NYT
CNN
Lack of M9s in record seemed consistent with model that M9s
only occur where lithosphere younger than 80 Myr subducts
faster than 50 mm/yr (Ruff and Kanamori, 1980)
Disproved by
Sumatra 2004
M9.3 and
dataset
reanalysis
(Stein & Okal,
2007)
Short record
at most SZs
didn’t include
larger
multisegment
ruptures
Stein & Okal, 2011
Tsunami radiates energy perpendicular to fault
Thus largest landward of highest slip patches
http://www.coastal.jp/tsunami2011/index.ph
p?FrontPage
http://www.geol.tsukuba.ac.jp/~
yagi-y/EQ/Tohoku/
Historical record of
large tsunamis
QuickTime™ and a
decompressor
are needed to see this picture.
NYT 4/20/11
NY Times 3/21/11
Maps fail because of
- bad physics (incorrect description of
earthquake processes)
-bad assumptions (mapmakers’ choice
of poorly known parameters)
- bad data (lacking, incomplete, or
underappreciated)
- bad luck (low probability events)
and combinations of these
Accuracy of hazard map prediction depends on accuracy of
answers assumed to hierarchy of four basic questions
Where will large earthquakes occur?
When will they occur?
How large will they be?
How strong will their shaking be?
Uncertainty & map failure result because these
are often poorly known
“A game of chance against nature, of which we still don't know all
the rules” (Lomnitz, 1989)
Where will large earthquakes occur?
When will large earthquakes occur?
How large will they be?
How strong will the shaking be?
Slow plate
boundary
Africa-Eurasia
convergence rate
varies smoothly
(5 mm/yr)
NUVEL-1
Argus, Gordon, DeMets &
Stein, 1989
Swafford & Stein, 2007
GSHAP 1999
Slow plate
boundary
Africa-Eurasia
convergence rate
varies smoothly
(5 mm/yr)
NUVEL-1
Argus, Gordon, DeMets &
Stein, 1989
2003
2004
M 6.4
Swafford & Stein, 2007
M 6.3
GSHAP 1999
2008 Wenchuan earthquake (Mw 7.9) was
not expected: map showed low hazard
USGS
Hazard map - assumed steady state - relied on lack
of recent seismicity
Didn’t use GPS data showing 1-2 mm/yr
Earthquakes prior to the 2008 Wenchuan event
Aftershocks of the Wenchuan event delineating the rupture zone
M. Liu
Long record needed to see real hazard
1933
M 7.3
1929
M 7.2
Swafford & Stein, 2007
Map depends greatly on
assumptions & thus has large
uncertainty
GSC
“Our glacial
loading model
suggests that
earthquakes may
occur anywhere
along the rifted
margin which has
been glaciated.”
Stein et al., 1979
1985
Concentrated
hazard bull's-eyes
at historic
earthquake sites
2005
Diffuse
hazard along
margin
Plate Boundary
Earthquakes
•Major fault loaded rapidly at
constant rate
•Earthquakes spatially focused
& temporally quasi-periodic
Past is fair predictor
Plate B
Plate A
Earthquakes at
different time
Intraplate Earthquakes
•Tectonic loading collectively
accommodated by a complex
system of interacting faults
•Loading rate on a given fault
is slow & may not be constant
•Earthquakes can cluster on a
fault for a while then shift
Past can be poor predictor
Stein, Liu & Wang 2009
New Madrid 1991: because paleoseismology shows
large events in 900 & 1450 AD, like those of 1811-12
GPS studies started, expecting to find strain
accumulating consistent with large events ~500 years
apart
Science, April 1999
We found little or no motion:
Seismicity migrates
Recent cluster transient,
possibly ending
Hazard overestimated
Similar behavior in other continental interiors
“Large continental interior earthquakes reactivate ancient
faults … geological studies indicate that earthquakes on these
faults tend to be temporally clustered and that recurrence
intervals are on the order of tens of thousands of years or
more.” (Crone et al., 2003)
Liu, Stein & Wang 2011
during the period
prior to the period
instrumental events
Earthquakes in North China
Beijing
Bohai Bay
Ordos
Plateau
1303 Hongtong
M 8.0
Weihi rift
Large events often pop up where there was little seismicity!
Liu, Stein & Wang 2011
during the period
prior to the period
instrumental events
Earthquakes in North China
Beijing
Bohai Bay
Ordos
Plateau
Weihi rift
1556 Huaxian
M 8.3
Large events often pop up where there was little seismicity!
Liu, Stein & Wang 2011
during the period
prior to the period
instrumental events
Earthquakes in North China
Beijing
Bohai Bay
Ordos
Plateau
Weihi rift
1668 Tancheng
M 8.5
Large events often pop up where there was little seismicity!
Liu, Stein & Wang 2011
during the period
prior to the period
instrumental events
Earthquakes in North China
1679 Sanhe
M 8.0
Beijing
Bohai Bay
Ordos
Plateau
Weihi rift
Large events often pop up where there was little seismicity!
Liu, Stein & Wang 2011
during the period
prior to the period
instrumental events
Earthquakes in North China
1975 Haicheng
M 7.3
Beijing
1976 TangshanBohai Bay
M 7.8
Ordos
Plateau
1966 Xingtai
M 7.2
Weihi rift
Large events often pop up where there was little seismicity!
No large (M>7) events ruptured the same
fault segment twice in past 2000 years
Historical
Instrumental
Weihi rift
In past 200 years, quakes migrated from Shanxi Graben to N. China Plain
Maps are like ‘Whack-a-mole’ - you wait for
the mole to come up where it went down,
but it’s likely to pop up somewhere else.
Where will large earthquakes occur?
When will large earthquakes occur?
How large will they be?
How strong will the shaking be?
EARTHQUAKE RECURRENCE IS HIGHLY VARIABLE
Sieh et al., 1989
Extend earthquake history with
paleoseismology
M>7 mean 132 yr s 105 yr
Estimated probability in 30 yrs 7-51%
Assumed probability of large earthquake & thus
hazard depend on recurrence model & position in
earthquake cycle
Time
dependent
predicts
lower until
~2/3 mean
recurrence
Results
depend on
both model
choice &
assumed
mean
recurrence
Hebden & Stein, 2008
Not clear which model works best where
%106
154%
2% in 50 yr
(1/2500 yr)
Effect
larger in
Memphis
Large
uncertainty
in maps
Where will large earthquakes occur?
When will large earthquakes occur?
How large will they be?
How strong will the shaking be?
Gutenberg-Richter relationship (1944):
log10 N = a -b M
N = number of earthquakes occurring ≥ M
a = activity rate (y-intercept)
b = slope (commonly called b-value)
M = Magnitude
Simple power-law distribution
controlled by the scale invariance
of earthquakes (e.g. Turcotte, 1997)
Useful for estimating the
recurrence of large
earthquakes from a given
fault or region
Mmax crucial for hazard
models
GUTENBERG-RICHTER RELATIONSHIP: INDIVIDUAL FAULTS
Wasatch
instrumental data
Characteristic
Basel, Switzerland
historical data
Uncharacteristic
paleoseismic data
paleoseismic data
Youngs & Coppersmith, 1985
Meghraoui et al., 2001
Largest events deviate in either direction, often when different data
mismatch
When more frequent than expected termed characteristic
earthquakes. Alternative are uncharacteristic earthquakes
These - at least in some cases - are artifacts of short history that
overpredict or underpredict hazard
SHORT HISTORY SIMULATIONS
10,000 synthetic earthquake histories for G-R relation with slope b=1
Gaussian recurrence times for M> 5, 6, 7
Various history lengths given in terms of Tav, mean recurrence for M>7
For histories = 0.5 Tav any M7 earthquakes appear characteristic, since
can’t observe fractions of earthquakes
Thus either overestimate rate of largest earthquakes or
underestimate Mmax
Stein & Newman 2004
SHORT SIMULATIONS
Often overestimate rate of largest earthquakes or underestimate Mmax
Where will large earthquakes occur?
When will large earthquakes occur?
How large will they be?
How strong will the shaking be?
Effects of assumed
ground motion
model trade off
with Mmax
Effect as large as one
magnitude unit
Frankel model, developed for
maps, predicts significantly
greater shaking for M > 7
Frankel M 7 similar to
Atkinson & Boore or
Toro M 8
Models use various
combinations, e.g. 1996
averaged Frankel & Toro
models; Atkinson & Boore
not used
Newman et al., 2001
PREDICTED
HAZARD ALSO
DEPENDS
GREATLY ON
180%
- Assumed
maximum
magnitude of largest
events
-Assumed ground
motion model
-Neither are well
known since large
earthquakes rare
Newman et al., 2001
275%
What to do
Realistically assess uncertainties and
present them candidly to allow users to
decide how much credence to place
Develop methods to objectively test
hazard maps and thus guide future
improvements
Global warming forecasts present uncertainties by showing
factor of 3 range of model predictions
IPCC 2007
Warming by 2099
The AOGCMs cannot sample the full range of possible warming, in particular because they do not include
uncertainties in the carbon cycle. In addition to the range derived directly from the AR4 multi-model
ensemble, Figure 10.29 depicts additional uncertainty estimates obtained from published probabilistic
methods using different types of models and observational constraints: the MAGICC SCM and the
BERN2.5CC coupled climate-carbon cycle EMIC tuned to different climate sensitivities and carbon cycle
settings, and the C4MIP coupled climate-carbon cycle models. Based on these results, the future increase in
global mean temperature is likely to fall within –40 to +60% of the multi-model AOGCM mean warming
simulated for each scenario. This range results from an expert judgement of the multiple lines of evidence
presented in Figure 10.29, and assumes that the models approximately capture the range of uncertainties in
the carbon cycle. The range is well constrained at the lower bound since climate sensitivity is better
constrained at the low end (see Box 10.2), and carbon cycle uncertainty only weakly affects the lower bound.
The upper bound is less certain as there is more variation across the different models and methods, partly
because carbon cycle feedback uncertainties are greater with larger warming.
In addition to comparing maps, comparing model predictions
shows the large uncertainties resulting from different
assumptions
Shows contributions to logic tree before subjective weighting
Testing analogy: evidence-based medicine
objectively evaluates widely used treatments
Although more than 650,000 arthroscopic knee surgeries at a
cost of roughly $5,000 each were being performed each year,
a controlled experiment showed that "the outcomes were no
better than a placebo procedure."
QuickTime™ and a
decompressor
are needed to see this picture.
Need objective criteria to test maps by comparison to
what happened after they were published.
One is to compare maximum
acceleration observed over the
years to that predicted by both
map and null hypotheses.
A simple null hypothesis is
regionally uniformly distributed
seismicity.
Japanese map seems to be doing
worse than this null hypothesis,
implying overparametrized model
Geller
2011
Detailed model of segments with 30 year probabilities
Off Sanriku-oki North ~M8
0.2 to 10%
Off Sanriku-oki Central~M7.7
80 to 90%
Off Miyagi ~M7.5
> 90%
Off Fukushima ~M7.4
7%
Off Ibaraki ~M6.7 – M7.2
90%
Expected Earthquake Sources
50 to 150 km segments
M7.5 to 8.2
(Headquarters for Earthquake Research Promotion)
Sanriku to Boso M8.2 (plate boundary)
20%
Sanriku to Boso M8.2 (Intraplate)
4-7%
J. Mori
Giant earthquake broke all of the segments
Expected Earthquake Sources
50 to 150 km segments
M6.7 to 8.2
(Headquarters for Earthquake Research Promotion)
2011 Tohoku Earthquake
450 km long fault, M 9.1
(Aftershock map from USGS)
J. Mori
Some testing challenges
1) Short time record: can in some cases be worked around.
For example, North China record probably has almost or all
M7s in 2000 years. Paleoseismology can go back even
further, with higher probability of missing some.
2) Subjective nature of hazard mapping, resulting from need
to chose faults, maximum magnitude, recurrence model,
and ground motion model. This precludes the traditional
method of developing a model from the first part of a time
series and testing how well it does in the later part. That
works if the model is "automatically" generated by some
rules (e.g. least squares, etc). In the earthquake case, this
can't be done easily because we know what happens in the
later part of the series.
Summary
- Hazard maps depend dramatically on unknown and
difficult-to-assess parameters and hence on the
mapmakers’ preconceptions
- thus have large uncertainties that are generally
underestimated and not communicated to public
- sometimes either underpredict hazard (too much
aseismic slip) in areas where large earthquakes occur
- or overpredict hazard (too much seismic slip)
Without objective testing, maps won’t improve &
seismology will keep having to explain away
embarrassing failures
Challenge: Users Want Predictions
Future Nobel Prize winner Kenneth Arrow served as a military
weather forecaster. As he described,
“my colleagues had the responsibility of preparing long-range
weather forecasts, i.e., for the following month. The
statisticians among us subjected these forecasts to verification
and found they differed in no way from chance. The
forecasters themselves were convinced and requested that the
forecasts be discontinued.
The reply read approximately: "The commanding general is
well aware that the forecasts are no good. However, he needs
them for planning purposes."
Gardner, D., Future Babble: Why Expert Predictions Fail - and Why We Believe
Them Anyway, 2010