Collaboratory for the Study of Earthquake Predictability

1
Collaboratory for the Study of Earthquake
Predictability

• 1st CSEP Workshop
• Mandalay Beach Resort
• June 7-8, 2006

2
Collaboratory for the Study of Earthquake
Predictability (CSEP)

• Earthquake prediction research is hampered by
  inadequate infrastructure for conducting
  scientific prediction experiments
• SCEC has received 1.2 million from the W. M.
  Keck Foundation for a 3-yr program to develop
  CSEP infrastructure
• Primary objective rigorous comparative testing
  of scientific prediction experiments spanning a
  variety of fault systems to study the physical
  basis for earthquake predictability
• CSEP will build on the RELM program and similar
  efforts among its international partners
• This first CSEP workshop will develop an initial
  project plan

3
New Interest in the Prediction Problem

• Motivated by
• better data from seismology, geodesy, and geology
• new knowledge of the physics of earthquake
  ruptures
• a more comprehensive understanding of how active
  faults systems actually work

• Promising developments include
• Better catalogs that incorporate smaller events,
  source mechanisms, and other information from
  high-performance seismic networks
• Detection of new types of signals
• slow precursors on mid-ocean ridge transform
  faults
• silent earthquakes in subduction zones and along
  the San Andreas fault
• periodic slow slip events and related episodes of
  harmonic tremor on the lower reaches of
  subduction megathrusts
• Improved models of static and dynamic stress
  interactions among faults and the effects of
  earthquake stress evolution on seismicity

4
Some Basic Points

• Probabilistic seismic hazard analysis (PSHA)
  provides the conceptual framework for
  physics-based predictions of earthquake effects

• Long-term forecasting of earthquake ruptures
  based on tectonic fault models provides the
  locations, magnitudes, and frequencies needed for
  many PSHA applications (e.g., building codes)
• After a century of research, we still cannot
  predict large earthquakes with the short-term
  reliability to needed to prepare communities for
  impending disasters
• Dialog on earthquake prediction among
  scientists, as well as between the public and the
  scientific community has become corrupted by
  the controversies surrounding this type of
  operational earthquake prediction

5
Three Definitions

• Earthquake predictability
• degree to which the future occurrence of
  earthquakes is encoded in the behavior of an
  active fault system
• Scientific earthquake prediction
• a testable hypothesis, usually stated in
  probabilistic terms, of the location, time, and
  size of fault ruptures
• Useful earthquake prediction
• advance warning of potentially destructive fault
  rupture precise and reliable enough to warrant
  actions to prepare communities

6
Three Questions

• Q1. How should scientific earthquake predictions
  be stated and tested?
• - How should prediction experiments be conducted
  and evaluated?
• Q2. What is the intrinsic predictability of the
  earthquake rupture process?
• - Are there coherent space-time structures in the
  chaotic evolution of active fault systems?
• Q3. Can knowledge of large-earthquake
  predictability be deployed as useful predictions?
• - Is operational earthquake prediction feasible?

7
Although earthquakes seem to strike out of the
blue, the furious energy that a quake releases
builds up for months and years beforehand in the
form of stresses within Earth's crust. At the
moment, forecasters have no direct way of seeing
these stresses or detecting when they reach
critically high levels. That may be changing,
however. Satellite technologies being developed
at NASA and elsewhere might be able to spot the
signs of an impending quake days or weeks before
it strikes, giving the public and emergency
planners time to prepare.
i.e., might answer Q3
8
Silver Bullet Approach

• Seeks useful, short-term earthquake predictions
  i.e., focuses on direct answer to Q3
• motivated by laboratory studies of rupture
  nucleation
• dominated research in the 1970s and 1980s
• Searches for signals diagnostic of approach to
  rupture, including
• foreshocks
• strain precursors
• electromagnetic precursors
• hydrologic changes
• geochemical signals
• animal behavior
• Has not thus far led to useful prediction
  methodologies

9
Brick-by-Brick Approach

• Focused on experimentation (Q1) and
  predictability (Q2), not operational prediction
  (Q3)
• Built on system-specific models of stress
  transfer and earthquake triggering
• Probabilistic prediction of earthquakes on
  multiple time scales, incorporating geologic and
  geodetic information, as well as seismicity data
• Steady efforts to understand and improve
  predictability, even if probability gains are
  small
• Demonstrates predictability by rigorous testing
  based on intercomparison of algorithms
• RELM program and its extension to a Collaboratory
  for the Study of Earthquake Predictability (CSEP)

10

• Southern California Earthquake Center
• Involves 500 scientists at 55 institutions
  worldwide
• Focuses on earthquake system science using
  Southern California as a natural laboratory
• Translates basic research into practical products
  for earthquake risk reduction

SCEC Focus Groups
11
SCEC3 Science Priority Objectives

• Improve the unified structural representation and
  employ it to develop system-level models for
  earthquake forecasting and ground motion
  prediction
• Develop an extended earthquake rupture forecast
  to drive physics-based PSHA
• Define slip rate and earthquake history of
  southern San Andreas fault system for last 2000
  years
• Determine the origin and evolution of on- and
  off-fault damage as a function of depth
• Test hypotheses for dynamic fault weakening
• Assess predictability of rupture extent and
  direction on major faults
• Investigate implications of geodetic/geologic
  rate discrepancies for earthquake forecasting
• Develop a system-level deformation and
  stress-evolution model for earthquake forecasting
• Map seismicity and source parameters in relation
  to known faults
• Develop a geodetic network processing system that
  will detect anomalous strain transients
• Test of scientific prediction hypotheses against
  reference models to understand the physical basis
  of earthquake predictability
• Predict broadband ground motions for a
  comprehensive set of large scenario earthquakes
• Develop pseudo-dynamic source models consistent
  with dynamic rupture models
• Determine the upper limits of extreme ground
  motion
• Investigate the upper frequency limit of
  deterministic ground motion predictions
• Validate earthquake simulations
• Collaborate with earthquake engineers to develop
  rupture-to-rafters simulation capability for
  physics-based risk analysis
• Prepare post-earthquake response

12
SCEC Computational Pathways
Standard seismic hazard analysis
1
Empirical models
Intensity Measures
Attenuation Relationship
Earthquake Rupture Forecast
Extended Earthquake Rupture Forecast
1
13
Scenario ShakeMaps for M 7.7 Southern San Andreas
Rupture Based on Empirical Attenuation
Relationships
Without soil basin effects
With soil basin effects
Abrahamson Silva (1997)
Field (2000)
OpenSHA Platform (Field et al. 2003)
14
Physics-based PSHA requires prediction of
directivity and other rupture parameters
(extended ERF)
Rupture direction NW?SE
Rupture direction NW?SE
Southernmost San Andreas M7.7 (Olsen et al. 2006)
15
Types of Earthquake Prediction

• Prediction vs. Forecasting
• Predictions identify periods of increased
  probability relative to long-term forecasts
• Time scale of prediction
• Long-term (decades to centuries) ? forecasts
• Intermediate-term (months to years)
• Short-term (seconds to weeks)
• Input basis
• Data-based
• Model-based
• Output basis
• probability-based
• alarm-based
• Retrospective vs. prospective

16
Official U. S. Earthquake PredictionUSGS
National Seismic Hazard Mapping Project (2002)

• Specifies the maximum shaking expected over a
  long period of time (typically 50 years)
• at all U.S. sites
• from all potential earthquake sources
• Rupture forecast is based on time-independent
  (Poisson) probabilities
• Ignores information about current state of the
  fault system

17
Recurrence intervals for San Andreas
earthquakes 1906 210 yr 1857 206 yr 1690 220
yr California Geological Survey (1996)
Time-Dependent Earthquake Forecasting
1906 M 7.8
2072 ?
Probability
1857 M 7.9
? 2153 (Time-independent)
1690 M 7.7
Year
18
Long-Term Prediction

• SCEC goal
• Time-dependent forecasts with better skill than
  the National Seismic Hazard Maps
• Methodology guided by fault-system models
• Paleoseismic models of fault rupture histories
• Requires correlation of slip history at multiple
  sites
• Stress evolution models
• Requires system-level models capable of
  predicting stress state
• Synthesis by new Working Group on California
  Earthquake Probabilities (WGCEP)
• SCEC / USGS / CGS project will develop a Uniform
  California Earthquake Rupture Forecast (UCERF) by
  2007

19
(No Transcript)
20
Short-Term Prediction

• SCEC goal
• Establish short-term reference predictions using
  earthquake triggering models
• Epidemic Type Aftershock Sequence (ETAS) models
• Statistics of aftershock sequences are well
  behaved
• Clustering of foreshocks, mainshocks, and
  aftershocks can be described by the same seismic
  triggering mechanism
• Builds on many previous studies
• Y. Ogata (1988, etc.)
• P. Reasenberg L. Jones (1989, etc.)
• A. Helmstetter D. Sornette (2002, etc.)

Northridge earthquake (M 6.7) Jan 17, 1994
21
ETAS Prediction of Short-Term Seismicity
ETAS prediction
Background rate
Observed seismicity
Retrospective daily ETAS predictions of Southern
California seismicity by Helmstetter et al. (2005)
22
Short-Term Earthquake Probability (STEP) Map
http//pasadena.wr.usgs.gov/step
23
SCEC Community Fault Model
A. Plesch and J. Shaw (2003)
24
Relocated Seismicity 1984-2002(Hauksson
Shearer, 2005)
25
Seismicity Rate vs. Fault OffsetSouthern
California catalogs (5 km h 15 km)
Strike-slip fault
y0
N y0 0.85
Epicenter
Powers Jordan (2005)
26
Aftershock StacksShearer (2004) catalog for
Southern California (5 km h 15 km)
Mainshock shift from fault traces
Results suggest a modified ETAS spatial kernel of
the form
Powers Jordan (2005)
27
Evaluation of ETAS Model

• The ETAS model provides a good first-order
  description of earthquake triggering
• Suitable as a reference model for short-term
  predictions
• In Southern California, short-term predictions of
  seismicity rate based on ETAS achieve probability
  gain factors gt10 relative to long-term Poisson
  models (Helmstetter et al., 2005)
• Gain decreases with magnitude threshold i.e.,
  little gain for large earthquakes
• Use of fault-based models may allow improvements
• Some regions, such as ridge transform faults,
  show anomalous statistics and more
  predictability relative to ETAS
• J. McGuire, M. Boettcher T. H. Jordan,
  Foreshock sequences and short-term earthquake
  predictability on East-Pacific Rise transform
  faults, Nature, 434, 457-461, 2005

28
Comparison of RTF Foreshock/Aftershock Statistics
with ETAS Prediction
Foreshocks per Mainshock
Aftershocks per Mainshock
29
Slow Earthquakes and Deep Tremor in the Cascadia
Subduction Zone
14 months
Do slow slip transients in subduction zones
trigger fast ruptures?
Miller (2004)
30
Unification Across Scales

• Earthquake systems have significant
  predictability across a range of scales
• Large-scale, long-term fault-based models
• Small-scale, short-term at least as good as ETAS
• Unification across scales requires a focus on
  intermediate-term predictability
• Physical basis in stress evolution and transfer
• Statistical basis in seismicity patterns
• Integration into fault-system models

31
Stress Transfer in the North Anatolian Fault
System
Stein et al., J. Geophys. Res., 1997
32
Seismicity Patterns Used in Intermediate-Term
Prediction
log frequency
log magnitude
Accelerating Seismicity
Long-Range Correlation
Large-Magnitude Enrichment
33
Accelerating Seismicity Before the 2004 M 9
Sumatra-Andaman Islands Earthquake
Mignan, Bowman King, in preparation, 2005
34
Problems in Assessing Earthquake Rupture
Forecasts and Prediction Experiments

• Scientific publications provide insufficient
  information for independent evaluation
• Active researchers are constantly tweaking their
  procedures, which become moving targets
• Difficult to find resources to conduct and
  evaluate long-term prediction experiments
• Data to evaluate prediction experiments are often
  improperly specified
• Standards are lacking for testing predictions
  against reference forecasts

35
SCEC/USGS Working Group for the Development
of Regional Earthquake Likelihood Models
Seismicity-based model (Gerstenberger others)
Simulation-based model (Ward)
Geodetic-based model (Jackson others)
36
RELM Testing Program

• RELM is developing a variety of earthquake
  forecasts and predictions for Southern California
• Both data-based and model-based
• T-RELM is testing probability-based models
  against each other using log-likelihood scoring

RELM Testing Facility at ETHZ
37
Collaboratory for the Study of Earthquake
Predictability (CSEP)

• Extension of T-RELM to a more set of diverse
  prediction experiments spanning a variety of
  fault systems
• 3-yr program of infrastructure development funded
  by W. M. Keck Foundation
• Continued collaboration within SCEC expanded
  international partnerships
• Goals
• 1. Reduce the controversies through a
  collaboratory infrastructure that can support a
  wide range of scientific prediction experiments
• 2. Promote rigorous research on earthquake
  predictability through the SCEC program and its
  global partnerships
• 3. Help government agencies assess the
  feasibility of earthquake prediction and the
  performance of proposed prediction algorithms

38
CSEP Objectives Design

• 1. Establish rigorous procedures for registering
  and evaluating prediction experiments
• 2. Construct community standards and protocols
  for comparative testing of predictions
• 3. Develop an infrastructure that allows groups
  of researchers to participate in prediction
  experiments
• 4. Provide access to authorized data sets and
  monitoring products for calibrating and testing
  prediction algorithms
• 5. Accommodate experiments involving fault
  systems in different geographic and tectonic
  environments

39
(b)
(a)
0
100
200 km
(c)
(d)
40
Measures of Success

• M1. Procedures established for all RELM
  experiments during 1st year and for alarm-based
  algorithms during 2nd year
• M2. Consensus on testing standards and protocols
  is endorsed by the agency committees during first
  2 years
• M3. CSEP is testing prospective prediction
  experiments, including all RELM experiments, in
  2nd year
• M4. CSEP is hosting prediction experiments from
  fault systems outside California in 3rd year
• M5. Public communication of CSEP activities is
  judged to be effective by the SCEC External
  Advisory Committee and agency committees

41
Project Timeline

• CSEP fully operational, hosting full range
  prediction experiments from U.S. and other
  countries

Complete main phase of collaboratory development
open collaboratory to other researchers
M5
M4
Register prototype prediction experiments into
CSEP, drawn from the RELM Project
M3
M2
M1a
M1b
2006
2007
2008
2009
42
(No Transcript)
43
Major Issues

• CSEP infrastructure development plan
• Science program
• New SCEC focus group
• Coordination with government agencies
• USGS/NEPEC
• OES/CGS/CEPEC
• Communication with the public
• International collaborations
• Sustainability

44
Questions to Answer at This Workshop

• How should CSEP be organized within SCEC?
• What criteria will be used to establish natural
  laboratories for prediction experiments?
• How will CSEP interact with similar efforts in
  other countries?
• Should CSEP include global prediction
  experiments?
• What types of data will be considered?
• How will these data be certified?
• What types of prediction experiments will CSEP
  accept?
• How will they be registered into the
  collaboratory?
• How will they be evaluated?
• What cyberinfrastructure is needed to support
  CSEP?
• What are our short-term objectives?

45
Agenda
Wednesday, June 7 830 am Session 1
Introduction 845 am CSEP in the context of
SCEC3 (T. Jordan) 915 am RELM testing program
(D. Schorlemmer) 945 am Long-term forecasting
and the WGCEP (N. Field) 1015 am NEPEC
activities (M. Blanpied) 1030 am
Break 1045 am Session 2 Prediction Hypotheses
Testing Methods I Discussion leader B.
Minster Reporter I.
Zaliapin Initial presentation (20 min) D.
Jackson 1200 noon Lunch 130 pm Session 3
Natural Laboratories Data Authentication
I Discussion Leader T. Tullis Reporter K.
Felzer Initial presentation (20 min) M.
Gerstenberger 330 pm Break 345 pm Session 4
Prediction Hypotheses Testing Methods II
Discussion Leader J. Rundle Reporter D.
Bowman Initial presentation (20 min) J. Zechar
530 pm Complementary drinks in the Surf
Room 700 pm Dinner
Thursday, June 8 830 am Session 5 Natural
Laboratories and Data Authentication
II Discussion Leader C. Sammis Reporter J.
McGuire Initial presentation (20 min) Y.
Kagan 1000 am Break 1030 am Session 6 CSEP
Design, Organization Participation
Discussion Leader W. Ellsworth
Reporter D. Schorlemmer Initial
presentation (20 min) P. Maechling 1200
noon Lunch 100 pm Session 7 Consensus
Statements Recommendations (10 min reports)
Prediction Hypotheses and
Testing Methods I. Zaliapin D. Bowman Natural
Laboratories and Data Authentication K. Feltzer
J. McGuire CSEP Design, Organization
Participation D. Schorlemmer 300 pm
Adjournment
46
Topics for Presentations
D. Jackson Overview of the prediction problem,
likelihood testing methods M. Gerstenberger
Prediction experiments in New Zealand and
California J. Zechar Alarm-based testing
methods, ROC and Molchan diagrams Y. Kagan
Global prediction experiments P. Maechling CSEP
hardware and software design issues
47
End
48
CSEP Project Budget
49
Collaboratory for the Study of Earthquake
Predictability(Phase 1)
CSEP Working Group Data Standards
Model Standards Testing Standards
Committee Committee
Committee
Global Data Service
Global Testing Project
New Zealand Natural Lab
California Natural Lab
European Union Natural Lab
NZ Data Service
CA Data Service
EU Data Service
NZ Testing Center
SCEC Testing Center
EU Testing Center
CSEP Grid
Cyberinfrastructure Committee
50
SCEC Testing Center (Phase 1)
SCEC/CSEP Working Group User
Technical Architecture CEO
Committee Committee
Committee
NEPEC CEPEC
SCEC/EFP Focus Group
Data Standards Model Standards
Testing Standards TWG
TWG TWG
Model Registration
Test Processing
Data Service
USGS/CISN

• Gridded RELMs ( 2)
• Short-term RELMs
• STEP
• ETAS
• Fault-based forecasts
• NSHM
• UCERF
• AMR
• Global models

Likelihood tests ROC/Molchan tests CT tests
Raw
ANSS Catalog
DP-tagged
Declustered
NSHMP Fault Database
Fault model
Test Posting
Other CSEP Testing Centers
CSEP Grid
Cyberinfrastructure TWG