Working Group on California Earthquake Probabilities

1
Working Group on California Earthquake
Probabilities (WGCEP) Development of a Uniform
California Earthquake Rupture Forecast (UCERF)
2
WGCEP Goals
To provide the California Earthquake Authority
(CEA) with a statewide, time-dependent ERF that
uses best available science and is endorsed by
the USGS, CGS, and SCEC, and is evaluated by
Scientific Review Panel (SRP) and CEPEC
NEPEC Coordinated with the next National Seismic
Hazard Mapping Program (NSHMP) time-independent
model CEA will use this to set earthquake
insurance rates (they want 5-year forecasts,
maybe 1-year in future)
3
UCERF Model Components
Fault Model(s)
(A)
Black Box
Deformation Model(s)
(B)
Black Box
(C)
Earthquake Rate Model(s)
Black Box
(D)
Earthquake Prob Model(s)
4
UCERF Model Components
Fault Model(s)
(A)
Black Box
Instrumental Qk Catalog
Deformation Model(s)
Fault Section Database
Historical Qk Catalog
(B)
Black Box
(C)
Earthquake Rate Model(s)
GPS Database
Black Box
(D)
Paleo Sites Database
Earthquake Prob Model(s)
5
Delivery Schedule

• February 8, 2006 (to CEA)
• UCERF 1.0
• S. SAF Assessment to CEA
• Aug 31, 2006 (to CEA)
• Fault Section Database 2.0
• Earthquake Rate Model 2.0 (preliminary for
  NSHMP)
• April 1, 2007 (to NSHMP)
• Final, reviewed Earthquake Rate Model
• (for use in 2007 NSHMP revision)
• September 30, 2007 (to CEA)
• UCERF 2.0 (reviewed by SRP and CEPEC)

UCERFs 3 later
6
Aug 31 Deliverables
i. Fault Section Database 2.0 will contain
the parameters describing a revised statewide set
of fault sections. This statewide set will be
sufficient for building Earthquake Rate Model
2.0. ii. Earthquake Rate Model 2.0 will
be the product delivered to the USGS as
USC-SCEC/USGS/CGS input to NSHMP(2007).

• Reality this will be a preliminary version of
  whats ultimately used, as revisions up to the
  last minute are inevitable (versions 2.X).
• NSHMP will have their own public review in
  October, 2006.
• Thus, how polished does the Aug. 31 delivery need
  to be?
• Branch weights will not be included (except maybe
  preliminary)

7
Aug 31 Deliverables
i. Fault Section Database 2.0 will contain
the parameters describing a revised statewide set
of fault sections. This statewide set will be
sufficient for building Earthquake Rate Model
2.0. ii. Earthquake Rate Model 2.0 will
be the product delivered to the USGS as
USC-SCEC/USGS/CGS input to NSHMP(2007).

• Incremental changes to NSHMP (2002) because
• We want to keep track of what changes matter
• Dont promise more than we can deliver
• NSHMP consumers dont want more than this
• (more ambitious changes in versions 3.0)

8
Aug 31 Deliverables
i. Fault Section Database 2.0 will contain
the parameters describing a revised statewide set
of fault sections. This statewide set will be
sufficient for building Earthquake Rate Model
2.0. ii. Earthquake Rate Model 2.0 will
be the product delivered to the USGS as
USC-SCEC/USGS/CGS input to NSHMP(2007).

• Development strategy
• Focus on whats important rather than whats
  interesting
• Statewide consistency
• Simplify wherever possible
• Question
• What hazard or loss metric should we use?

9
Time Span
Earthquake- Rupture Forecast List of
Adjustable Parameters
Intensity- Measure Relationship List of
Supported Intensity-Measure Types List of
Site-Related Independent Parameters
Site Location List of Site- Related Parameters
Intensity Measure Type Level (IMT IML)
Adjustable Parameter Settings
Hazard Calculation
Prob(IMTIML)
10
UCERF Model Components
Fault Model(s)
(A)
Black Box
Deformation Model(s)
(B)
Black Box
(C)
Earthquake Rate Model(s)
Black Box
This is a time- independent ERF
(D)
Earthquake Prob Model(s)
11
The ERF Adjustable Parameters are the epistemic
uncertainties
12
Those in Earthquake Rate Model 2.0 include (so
far)
13

• Easily Added
• Fault Section Database 2.0 Uncertainties for
• slip rate
• upper and lower seismogenic depth
• aseismicity factors (if available)
• average dip for each fault section
• Fraction of Mo-Rate on A B sources into smaller
  events
• Additional epistemic uncertainty on A- B-fault
  mags
• More Work
• Other slip per even assumptions on A-Faults
• (other than the characteristic slip Dsr Ds)
• WG02 Slip (Dsr proportional to vs)

These could (should?) be added to Earthquake Rate
Model 2.0
14
Definitely Not Included in Earthquake Rate Model
2.0
Deformation Models 3.0 Fault-to-fault rupture
jumps (via generalized inverse or
simulations) Relaxation of assumed segmentation
while honoring paleoseismic data (which might
demand a segmented model)
These inherent limits should be kept in mind when
deciding how much more work should be put into
Earthquake Rate Models 2.x
15
The outer branches of a logic tree represent an
ERF Epistemic List (a list of ERFs w/ diff.
param. settings associated weights)
How extensively should the logic-tree be
sampled? Example with WGCEP-2002 What will
actually be used?
16
Basic Questions for SRP
What more should be accomplished by Aug.
31st? What should be accomplished by April 1
(NSHMP deadline)? What hazard or loss metric
should be used? How extensively should the
logic-tree be sampled? How shall the branch
weights be assigned? How shall the formal review
proceed?
Fault Section Database 2.0 Fault Models 2.1
2.2 Deformation Models 2.x Earthquake
Catalog Regional Seismicity Constraints Magnitude-
Area Relations Segment Recurrence Data Alt.
A-Fault Rupture Models Type-B Fault C-zone
models Recipe for combining everything
17
(No Transcript)
18
Intro to Fault Section Database, Fault Models,
and Deformation Models
19
(No Transcript)
20
Hanks Bakun (2002) M 3.98 log(A) if
Alt537 M 3.07 (4/3)log(A) if Agt537
Ellsworth A M 4.1 log(A)
Ellsworth B M 4.2 log(A)
Somerville (2006) M 3.98 log(A)
21
(No Transcript)
22
Type-A Fault Rupture Models
If S segments, then RS(S1)/2 different ruptures
involving contiguous segments. We want the
long-term rate (fr) of each rth rupture. We know
for each segment Slip Rate (vs) Mean Recur Int
(Ts1/?s) Constraints are Equation Set
(1) Equation Set (2) Equation Set
(3) Positivity where Dsr is the average
slip in the rth rupture on the sth segment, and
Gsr is a matrix indicating whether the rth
rupture involves the sth segment (1 if so, 0 if
not).

• Under-determined
• (infinite number of solutions)

23
WGCEP-2002 Solution
Requires moment balancing
Implicitly assumes Drs is proportional to vs
If end segment cant go alone, then neither can
its neighbor.
24
Current WGCEP Solution
Assume characteristic slip (like WGCEP-1995)
Dsr Ds vs/?s vsTs
With compiled Ts (spreadsheet), and vs from the
chosen deformation model, solve for the following
(by hand)

• Minimum Rate Solution - That which minimizes the
  total rate of ruptures (and therefore maximizes
  event magnitudes), consistent w/ obs.
• Maximum Rate Solution - That which maximizes
  the total rate of ruptures (and therefore
  minimizes event magnitudes), consistent w/ obs.
• Geological Insight Solution - That which makes
  all fr as close as possible to a complete set of
  defined by geologists, consistent w/ obs.

Equal Rate Solution - That which makes all fr as
equal as possible
(These dont span solution space, but should span
hazard/loss space?)
25
Current WGCEP Solution
Further details
Aseismic Slip Factor applied as reduction of area
or slip rate. Each rupture given a Gaussian
magnitude PDF w/ default sigma 0.12 and
truncation at /- 2 sigma.
26
Current WGCEP Solution
Hanks Bakun (2002) M 3.98 log(A) if
Alt537 M 3.07 (4/3)log(A) if Agt537
Ellsworth A M 4.1 log(A)
Ellsworth B M 4.2 log(A)
Somerville (2006) M 3.98 log(A)
27
Current WGCEP Solution
What needs further consideration (?)
Dependence of tabulated mean recurrence intervals
(Ts) on deformation model (vs). Some rupture
models are not exactly rate balanced. Influence
of mean recurrence interval uncertainties. How
well consistently are these uncertainties
defined? Alternatives to the characteristic slip
assumption (currently trying to implement a
generalized inversion solution using NNLS of
Lawson and Hanson (1974)).
28
Current WGCEP Solution
We also have an Un-Segmented option for Type-A
faults (same as for Type-B faults).
29
(No Transcript)
30
Type-B Fault Rupture Models
One for each fault section in the database that
is not part of a Type-A fault (although
Concord-Green Valley Greenville are sections
combined

• Set the following
• 1) Deformation Model
• 2) Aseismicity Factor Reduced Area?
• Mag-Area Relationship (to get mean and
• upper mag for char and GR dists)
• 4) Char vs GR
• 5) Mag Sigma (for char dist)
• 6) Truncation Level (for char dist)
• 7) B-Faults b-value (for GR dist)

• GR lower mag 6.5. If GR upper mag lt 6.5 , all
  moment-rate goes in the Char dist.
• NSHMP-2002 used a truncation level of 1.25 sigma
  (rather than the 2 sigma of WGCEP-2002), but also
  added an additional /- 0.2 epistemic uncertainty
  to both the Char and Upper-GR Mags (not yet done
  here).

31
Type-B Fault Rupture Models
NSHMP-2002 B-Fault exceptions (special cases with
fixed values)
Owl Lake ( M 6.5, rate 0.002/yr), Owens Valley
fault (M7.6, rate 0.00025) The magnitude was
fixed at the magnitude of the 1882? (1872)
earthquake. Honey Lake fault (M 6.9, rate
0.00067) recurrence rate of about 1500 years
based on paleoseismic study by Wills and
Borchardt (1991)   Eureka Peak ( M 6.4, rate
0.0002) The 5,000 yr recurrence is similar to the
other Mojave faults Burnt Mountain (M6.5, rate
0.0002) The 5,000 yr recurrence is the same as
other Mojave faults Cucamonga (M6.9, 0.00154)
Maximum magnitude and recurrence of about 650
years based on 2 m average recurrence. Sierra
Madre-San Fernando (M6.7, 0.001) Magnitude based
on San Fernando earthquake, recurrence of 1 ka
based on USGS (1996). Palos Verdes (M7.2,
0.00154) Recurrence based on McNeilan et al.
(1996) of 650 years. Blackwater (M 6.9, 0.0002)
Based on paleoseismic studies by Hecker et al.,
1993 Rubin and Sieh, 1993 and Herzberg and
Rockwell, 1993. Calico-Hidalgo (M7.2, 0.0002)
Based on paleoseismic studies by Hecker et al.,
1993 Rubin and Sieh, 1993 and Herzberg and
Rockwell, 1993. Gravel Hills-Harper Lake (M7.0,
0.0002) Based on paleoseismic studies by Hecker
et al., 1993 Rubin and Sieh, 1993 and Herzberg
and Rockwell, 1993. Helendale-S. Lockhart (M7.2,
0.0002) Based on paleoseismic studies by Hecker
et al., 1993 Rubin and Sieh, 1993 and Herzberg
and Rockwell, 1993. Lenwood-Lockhart-Old Woman
Springs (M7.5, 0.0002) Based on paleoseismic
studies by Hecker et al., 1993 Rubin and Sieh,
1993 and Herzberg and Rockwell,
1993. Pisgah-Bullion Mountain Mesquite Lake
(M7.2, 0.0002) Based on paleoseismic studies by
Hecker et al., 1993 Rubin and Sieh, 1993 and
Herzberg and Rockwell, 1993. Landers (M7.3,
0.0002) Based on paleoseismic studies by Hecker
et al., 1993 Rubin and Sieh, 1993 and Herzberg
and Rockwell, 1993. Magnitude based on 1992
Landers qk South Emerson-Copper Mountain (M 6.9,
0.0002) Based on paleoseismic studies by Hecker
et al., 1993 Rubin and Sieh, 1993 and Herzberg
and Rockwell, 1993. Johnson Valley (M 6.7,
0.0002) Based on paleoseismic studies by Hecker
et al., 1993 Rubin and Sieh, 1993 and Herzberg
and Rockwell, 1993. Maacama (M 7.5 ) Floated a M
7.5 earthquake because the discontinuous strands
of the fault were not thought to be capable of
rupturing in a larger event.
32
Type-C Zone Rupture Models
For events in areas where deformation is
occurring over a wide area on poorly known or
unknown faults, modeled as strike slip events
oriented along the structural trend (fixed
strike) GR between mag 6.5 and 7.3 (except
foothills gets mag 6-7) moment rate as
?LhslipRate.
4 New zones
33
(No Transcript)
34
Background Seismicity

• In computing total regional seismicity
  parameters, Karen recommends several revisions to
  the NSHMP-2002 method, including
• making corrections for magnitude error and
  rounding before calculating a values
• using only modern instrumental data to calculate
  b value
• leaving aftershocks in the catalog when
  calculating parameters for the entire study
  region
• using an inverse power law rather than a Gaussian
  smoothing kernel when calculating spatially
  variable seismicity rates

• Results
• 14 lower total rate of M5 events
• b-value of 1.0 rather than 0.8
• a different spatial distribution or seismicity

35
Background Seismicity

• In computing total regional seismicity
  parameters, Karen recommends several revisions to
  the NSHMP-2002 method, including
• making corrections for magnitude error and
  rounding before calculating a values
• using only modern instrumental data to calculate
  b value
• leaving aftershocks in the catalog when
  calculating parameters for the entire study
  region
• using an inverse power law rather than a Gaussian
  smoothing kernel when calculating spatially
  variable seismicity rates

• Results
• 14 lower total rate of M5 events
• b-value of 1.0 rather than 0.8
• a different spatial distribution or seismicity
• Her best estimate of the total rate of M5 events
  is 10/-1.2 per including aftershocks , and
  5.4/- 0.85 otherwise (and assuming the
  definition of aftershocks applied by the
  NSHMP-2002). However, for our definition of
  California (RELM-test region) mult these by
  0.75 to get
• Rate of M5 7.5 /- 0.9 including aftershocks
• Rate of M5 4.0 /- 0.6 excluding aftershocks

36
Background Seismicity
Background seismicity is computed as the total
target rate minus the rates implied by the
Type-A, TypeB, and C-zone sources. The
adjustable parameters (epistemic uncertainties)
are Total Rate of M5 events (default7.5
including aftershocks 4.0 excluding) Regional
b-value (default1.0) Maximum Magnitude of
Background (default7 as in NSHMP) The magnitude
frequency distribution for all background
seismicity combined is computed as follows 1)
Create a target cumulative Gutenberg-Richter
distribution between magnitude 5.0 and the
maximum magnitude of the background seismicity
(truncated on the incremental distribution) with
the specified b-value and total rate of M5
events. 2) From this target distribution
subtract the cumulative magnitude frequency
distribution of all the Type-A, Type-B, and
C-zone sources. 3) Set any negative rates in the
resultant magnitude-frequency distribution to
zero. What remains is then applied as background
seismicity using the relative spatial
distribution given in Appendix G (and taking care
to lower the relative rates accordingly over the
A-Fault, B-fault, and C-zone sources). Total
regional moment rate of model varies with the
assumed max-mag of background.
37
(No Transcript)
38
Results
Magnitude frequency distributions (MFDs) obtained
from Earthquake Rate Model 2.0 with default
parameters (listed in Table 1 and shown in Figure
3). The bold black line is the total model MFD,
blue is for all Type A sources, green is for the
Gutenberg-Richter part of all Type B sources,
charcoal is for the characteristic part of all
Type B sources, and hot pink is for the
background sources. Red is simply a target MFD
that should be ignored at high magnitudes.
39
Results
Changing Mag-Area Relationship
Ellsworth-B Regional b-value 0.9.
40
(No Transcript)
41
Those in Earthquake Rate Model 2.0 include (so
far)
42

• Easily Added
• Fault Section Database 2.0 Uncertainties for
• slip rate
• upper and lower seismogenic depth
• aseismicity factors (if available)
• average dip for each fault section
• Fraction of Mo-Rate on A B sources into smaller
  events
• Additional epistemic uncertainty on A- B-fault
  mags
• More Work
• Other slip per even assumptions on A-Faults
• (other than the characteristic slip Dsr Ds)
• WG02 Slip (Dsr proportional to vs)

These could (should?) be added to Earthquake Rate
Model 2.0
43
Definitely Not Included in Earthquake Rate Model
2.0
Deformation Models 3.0 Fault-to-fault rupture
jumps (via generalized inverse or
simulations) Relaxation of assumed segmentation
while honoring paleoseismic data (which might
demand a segmented model)
These inherent limits should be kept in mind when
deciding how much more work should be put into
Earthquake Rate Models 2.x
44
The outer branches of a logic tree represent an
ERF Epistemic List (a list of ERFs w/ diff.
param. settings associated weights)
How extensively should the logic-tree be
sampled? Example with WGCEP-2002 What will
actually be used?
45
Basic Questions for SRP
What more should be accomplished by Aug.
31st? What should be accomplished by April 1
(NSHMP deadline)? What hazard or loss metric
should be used? How extensively should the
logic-tree be sampled? How shall the branch
weights be assigned? How shall the formal review
proceed?
Fault Section Database 2.0 Fault Models 2.1
2.2 Deformation Models 2.x Earthquake
Catalog Regional Seismicity Constraints Magnitude-
Area Relations Segment Recurrence Data Alt.
A-Fault Rupture Models Type-B Fault C-zone
models Recipe for combining everything
46
(No Transcript)
47
(No Transcript)
48
(No Transcript)
49
SCEC will provide CEA with a single-point
interface to the project.
WGCEP Organization Funding Sources
CEA
Geoscience organizations
SCEC
NSF
Management oversight committee
Scientific review panel
USGS Menlo Park
USGS
MOC
SRP
Sources of WGCEP funding
USGS Golden
CGS
State of CA
WGCEP ExCom
Working Group on California Earthquake
Probabilities
Working group leadership

Subcom. A
Subcom. B
Subcom. C

Task-oriented subcommittees
50
WGCEP Management

• WGCEP Management Oversight Committee (MOC)
• SCEC Thomas H. Jordan (CEA contact)
• ? USGS, Menlo Park Rufus Catchings
• USGS, Golden Jill McCarthy
• CGS Michael Reichle

In charge of resource allocation and approving
all project plans, budgets, and schedules Their
signoff will constitute the SCEC/USGS/CGS
endorsement
51

• WGCEP Executive Committee
• Edward (Ned) Field SCEC/USGS, Pasadena
• ? Thomas Parsons, USGS, Menlo Park
• Chris Wills, CGS
• Ray Weldon, SCEC/UofO
• Mark Petersen, USGS, Golden
• Ross Stein, USGS, Menlo Park

Responsible for convening experts, reviewing
options, making decisions, and orchestrating
implementation of the model and supporting
databases Role of leadership is not to advocate
models, but to accommodate whatever models are
appropriate
Key Scientists Provide expert opinion and/or
specific model elements - likely receiving
funding documenting their contributions.
Contributors
52

• Scientific Review Panel
• Bill Ellsworth (chair)
• Art Frankel
• David Jackson
• Jim Dieterich
• Lloyd Cluff
• Allin Cornell
• Mike Blanpied
• David Schwartz

This group will ultimately decide whether weve
chosen a minimum set of alternative models that
adequately spans the range of viable 5-year
forecasts for California
CEPEC Lucile Jones Duncan Agnew Tom
Jordan Mike Reichle Jim Brune David
Openheimer William Lettis Paul Segall John
Parrish
53
Issues/Possible Innovations

• Statewide model
• Use of CFM (including alternatives)
• Use GPS data via kinematically consistent
  deformation model(s)
• Relax strict segmentation assumptions
• Allow fault-to-fault jumps
• Apply elastic-rebound-motivated renewal models in
  (4) (5)
• Include earthquake triggering effects
• Deploy as extensible, adaptive (living) model
• Simulation enabled

54
Decision Making Process
Two type of decisions 1) what model
components to include (logic-tree branches ) 2)
what weights to apply to each
Decisions will be made and a case-by-case (or
branch-by-branch) basis (web site has details
www.WGCEP.org).
55
Decision Making Process
In general 1. The ExCom hosts meetings/workshops
to solicit expert opinion. 2. The ExCom, with
perhaps assistance from others, drafts proposed
branches and preliminary weights with full
documentation and posts these on the web. 3.
Email feedback is requested from the broader
community and responses are entered into an
official record. 4. The ExCom revises and
documents accordingly. 5. The SRP reviews the
entire process and iterates with the ExCom if
need be (MOC serves as referee).
56
Decision Making Process
This entire decision making process will be well
documented for posterity. We will also strive
to establish a quantitative basis for setting
weights, both for numerical reproducibility and
future modifications. However, it may be that
"gut feeling" will in some cases be the best or
only way to assimilate a large number of
constraints.
57
Validation Verification
Verification will be conducted via standard
practice in software development (e.g., JUnit
Testing for our Java Classes). Validation via
participation in RELM/CSEP testing efforts
(although these wont be definitive
anytime). Test the assumptions that go into the
models. Examine simulated catalogs.
Both validation and verification will be
addressed on a case-by-case basis we will have
explicit sections dedicated to each in the formal
documentation of all model components.
58
More Info?
UCERF 1 vs UCERF 2 UCERF 2 Logic Tree Possible
Innovations

• Statewide model
• Use of CFM (including alternatives)
• Use GPS data via kinematically consistent
  deformation model(s)
• Relax strict segmentation assumptions
• Allow fault-to-fault jumps
• Apply elastic-rebound-motivated renewal models in
  (4) (5)
• Include earthquake triggering effects
• Deploy as extensible, adaptive (living) model.
• Simulation enabled

59
UCERF 1 vs 2

• Updated/revised fault models, slip rates and
  aseismic-slip-factor estimates
• Revision of rupture models for type-A faults
  based on new information, and to achieve more
  statewide consistency with respect to the range
  of segmented vs cascade vs floating-rupture
  models.
• Reexamination of type B-faults and their
  magnitude-frequency distributions
• Reconsideration of how historical seismicity is
  smoothed to generate the distribution of
  background events
• Apply the range of time dependent probability
  models considered by WGCEP-2002 on a consistent,
  statewide basis (making adjustments/ improvements
  where necessary)

60
Issues/Possible Innovations

• Statewide model
• Use of CFM (including alternatives)
• Use GPS data via kinematically consistent
  deformation model(s)
• Relax strict segmentation assumptions
• Allow fault-to-fault jumps
• Apply elastic-rebound-motivated renewal models in
  (4) (5)
• Include earthquake triggering effects
• Deploy as extensible, adaptive (living) model.
• Simulation enabled

61
Issues/Possible Innovations
1) Statewide model
62
Issues/Possible Innovations
2) Use of CFM (including alternatives)
take this state wide?
63
Issues/Possible Innovations
3) Use GPS data via kinematically consistent
deformation model(s)
64
Fault Models(s)
Black Box
Deformation Model(s)
We want 1) Improved slip rates on major
faults 2) Strain rates elsewhere 3) GPS data
included
Black Box
Earthquake Rate Model(s)
4) Kinematically consistent 5) Accommodate
all important faults 6) Can accommodate
alternative fault models 7) Accounts for
geologic and geodetic data uncertainties 8)
Includes viscoelastic effects 9) Includes
significant 3D effects 10) Statewide application
Black Box
Earthquake Prob Model(s)
65
No model has all these attributes
Fault Models(s)
Are any existing models better than sticking to
what we have? WGCEP recommending pursuit of
Black Box
Deformation Model(s)
NeoKinema Harvard-MIT Block Model Parsons
FEM Shen Zeng Perhaps others

• We want
• 1) Improved slip rates on major faults
• 2) Deformation rates elsewhere
• 3) GPS data included
• 4) Kinematically consistent
• 5) Accommodate all important faults
• 6) Can accommodate alternative fault models
• 7) Accounts for geologic and geodetic data
  uncertainties
• 8) Includes viscoelastic effects
• 9) Includes significant 3D effects
• 10) Statewide application

Black Box
Earthquake Rate Model(s)
Black Box
Earthquake Prob Model(s)
66
No model has all these attributes Are any
existing models better than doing nothing? WGCEP
recommending pursuit of
Fault Models(s)
Black Box
Deformation Model(s)
NeoKinema Harvard-MIT Block Model Parsons
FEM Perhaps others?
Black Box
Delivery will be revised slip rates on modeled
faults (and perhaps deformation rates elsewhere,
and stressing rates on faults)
Earthquake Rate Model(s)
Black Box
Earthquake Prob Model(s)
67
Issues/Possible Innovations
4) Relax strict segmentation
68
Issues/Possible Innovations
4) Relax strict segmentation
But ...
69
Issues/Possible Innovations
4) Relax strict segmentation
Does it matter (all models are discretized to
some extent)?
70
Issues/Possible Innovations
5) Allow fault-to-fault jumps
No previous WGCEPs have allowed such ruptures.
71
Issues/Possible Innovations
5) Allow fault-to-fault jumps
But ...
72
Issues/Possible Innovations
5) Allow fault-to-fault jumps
Fault Interactions and Large Complex Earthquakes
in the Los Angeles Area Anderson, Aagaard,
Hudnut (2003, Science 320, 1946-1949) We find
that a large northern San Jacinto fault
earthquake could trigger a cascading rupture of
the Sierra Madre-Cucamonga system, potentially
causing a moment magnitude 7.5 to 7.8 earthquake
on the edge of the Los Angeles metropolitan
region.
73
Issues/Possible Innovations
5) Allow fault-to-fault jumps
Can dynamic rupture modelers help define
fault-to-fault jumping probabilities?
74
Issues/Possible Innovations
6) Figure out how to apply elastic-rebound-motivat
ed renewal models properly
Problem how to compute conditional
time-dependent probabilities when you allow both
single and multi-segment ruptures (let alone
relaxing segmentation)? The way previous WGCEPs
have modeled this seems logically inconsistent
75
From WGCEP-2002
76
From WGCEP-2002
77
How then do we compute conditional probabilities
where we have single and multi-segment ruptures?
We have ideas
78
Issues/Possible Innovations
7) Include earthquake triggering effects
CEA wants 1- to 5-year forecasts
This is between the 30-year forecasts of
previous WGCEPs (renewal models) and the 24-hour
forecasts of STEP (the CEPEC-endorsed model based
on aftershock statistics)
We are attempting to design a framework that
could accommodate a variety of alternative
approaches (e.g., from RELM)
79
Issues/Possible Innovations
8) Deploy as extensible, adaptive (living) model.
i.e., modifications can be made as warranted by
scientific developments, the collection of new
data, or following the occurrence of significant
earthquakes. The model can be living to the
extent that update evaluation process can occur
in short order. CEA wants this.
80
Issues/Possible Innovations
9) Simulation enabled (i.e, can generate
synthetic catalogs)
Needed to go beyond next-event forecasts if
stress interactions and earthquake triggering are
included Helpful (if not required) to understand
model behavior Can be used to calibrate the
model (e.g., moment in aftershocks) If we can
deal with simulated events, well be ready for
any real events
81
Implementation Plan

• Guiding principles
• If it aint broke, dont fix it
• Some of the hoped-for innovations wont work out
• Everything will take longer than we think
• Build components in parallel (not in series)
• Get a basic version of each component implemented
  ASAP, and add improved versions when available

We cannot miss the NSHMP and CEA delivery
deadlines!
82
(No Transcript)
83
UCERF 1.0
Fault Model 1.0
NSHMP-2002 Fault Model
By Feb. 8, 2006
Black Box
Deformation Model 1.0
NSHMP-2002 Fault Slip Rates
NSHMP-2002 Earthquake Rate Model
Black Box
Earthquake Rate Model 1.0
Black Box
Earthquake Prob Model 1.0
Simple conditional probabilities based on date of
last earthquakes
84
UCERF 2.0
Fault Models 2.X
Revision of NSHMP-2002 Fault Model based on SCEC
CFM (including alternatives) any desired
changes for N. California
By Sept. 30, 2007
Black Box
Deformation Model 2.0
Revised slip rates for elements in Fault Model
2.0, perhaps constrained by GPS data.
New model based on reevaluation of fault
segmentation and cascades (e.g., based on Weldon
et al.) may relax segmentation and allow
fault-to-fault jumps.
Black Box
Earthquake Rate Model 2.0
Black Box
Earthquake Prob Model 2.0
Application of more sophisticated time- dependent
probability calculations
85
UCERF 3.0
Fault Models 2.X
Black Box
Deformation Model 3.0
e.g., relax segmentation, allow fault-to-fault
jumps, and use off-fault deformation rates to
help constrain off-fault seismicity.
Black Box
Earthquake Rate Model 3.0
Black Box
Earthquake Prob Model 3.0
e.g., enable real-time modification of
probabilities base on stress or seismicity-rate
changes.
86
UCERF 3.0
Fault Models 2.X
Black Box
Deformation Model 3.0
Use of a more sophisticated, California- wide
deformation model such as NeoKinema.
e.g., relax segmentation, allow fault-to-fault
jumps, and use off-fault deformation rates to
help constrain off-fault seismicity.
Black Box
Earthquake Rate Model 3.0
All of these are relatively ambitious and
delivery cannot be guaranteed by 2007.
Black Box
Earthquake Prob Model 3.0
e.g., enable real-time modification of
probabilities base on stress or seismicity-rate
changes.
87
(No Transcript)
88
UCERF Model Components
Fault Model(s)
(A)
Black Box
Deformation Model(s)
(B)
Black Box
(C)
Earthquake Rate Model(s)
Black Box
(D)
Earthquake Prob Model(s)
89
UCERF Model Components
Fault Model(s)
(A)
Black Box
Instrumental Qk Catalog
Deformation Model(s)
Fault Section Database
Historical Qk Catalog
(B)
Black Box
(C)
Earthquake Rate Model(s)
GPS Database
Black Box
(D)
Paleo Sites Database
Earthquake Prob Model(s)
90
UCERF Model Components
Fault Model(s)
(A)
Black Box
Instrumental Qk Catalog
Deformation Model(s)
Fault Section Database
Historical Qk Catalog
(B)
Black Box
(C)
Earthquake Rate Model(s)
GPS Database
Black Box
(D)
Paleo Sites Database
Earthquake Prob Model(s)
91
UCERF Model Components
Fault Model(s)
(A)
Black Box
Instrumental Qk Catalog
Deformation Model(s)
Fault Section Database
Historical Qk Catalog
(B)
Black Box
(C)
Earthquake Rate Model(s)
GPS Database
Black Box
(D)
Paleo Sites Database
Earthquake Prob Model(s)
92
UCERF Model Components
Fault Model(s)
(A)
Black Box
Instrumental Qk Catalog
Deformation Model(s)
Fault Section Database
Historical Qk Catalog
(B)
Black Box
(C)
Earthquake Rate Model(s)
GPS Database
Black Box
(D)
Paleo Sites Database
Earthquake Prob Model(s)
93
UCERF Model Components
Fault Model(s)
(A)
Black Box
Instrumental Qk Catalog
Deformation Model(s)
Fault Section Database
Historical Qk Catalog
(B)
Black Box
(C)
Earthquake Rate Model(s)
GPS Database
Black Box
(D)
Paleo Sites Database
Earthquake Prob Model(s)