Transcript Document

EarthScope
Knowledge
Know
Transfer Development
in the Context of
Seismic Hazard Analysis (SHA)
Edward (Ned) Field
USGS, Pasadena
Themes of this Talk
• SHA needs more physics
• SHA needs Information Technology
• Will EarthScope help SHA?
Short Version of Talk
Improved Seismic Hazard Analysis:
• requires a more physics based approach, which …
• requires system level, multidisciplinary effort, which …
• requires modularization of the various contributions
(community models and databases), which …
• will need interoperability, which …
• we don’t know how to achieve, which …
• is why we need and have an IT collaboration.
• EarthScope will contribute to long-term SHA goals
(but could hinder as well).
Seismic Hazard Analysis
Two Components:
(1) Earthquake Forecast
Probability in time and
space of all M≥5 events
(2) Ground-Motion Estimation
Intensity Measure
Regressions
(attenuation relations)
Full
waveform
modeling
Part I
SHA needs more physics
Seismic Hazard Analysis
(probabilistic)
Two Components:
(1) Earthquake Forecast
Probability in time and
space of all M≥5 events
(2) Ground-Motion Estimation
Intensity Measure
Relations
(attenuation relations)
Full
waveform
modeling
SCEC Phase-III Report
On the extent to which site effects can
be accounted for in SHA
14 papers
New Regressions:
An important Phase-III Conclusion:
Prediction uncertainty
remains high on empirical
regressions even after all
possible site-effect
corrections have been made
(surprising to engineers).
IMPLICATION: Statistical regression approach
appears to be reaching a point of diminishing returns
More accurate ground-motion estimates
will apparently require waveform
modeling from 1st principles of physics
SCEC Phase-III Report
Included 3D waveform simulations
14 papers
Magistrale
Olsen
Although 3D Waveform Simulations
constitute our best hope for more
accurate ground-motion predictions,
Unfortunately they are not yet accurate
enough for official SHA, nor
computationally fast enough for routine
use.
Improvements in
Waveform-modeling
will require:
(1) Improved structural representations
(2) Improved modeling of high frequencies,
scattering, and nonlinear effects
(3) Carefully orchestrated validation exercises
(4) Propagation of all significant uncertainties
Note:
SHA will continue to rely on
intensity-measure regressions for
some time (improvements are
needed, and engineers have other
types they would like developed).
Seismic Hazard Analysis
Two Components:
(1) Earthquake Forecast
Probability in time and
space of all M≥5 events
(2) Ground-Motion Estimation
Intensity Measure
Relations
(attenuation relations)
Full
waveform
modeling
Current Status
The earthquake-forecast model applied in the National
Hazard maps uses a time-independent (Poisson)
recurrence model for each potential earthquake,
meaning each event is completely independent of all
other events (even those on the same fault).
The 2001 Fall AGU meeting had more than 100
abstracts on stress interaction effects and timedependent earthquake probabilities (lots of good ideas).
There is consensus that some form of time
dependence and stress interaction exists, but no
consensus on how to model it.
RELM
A Working Group
for the Development of
Regional Earthquake Likelihood Models
(Earthquake Forecasts)
A
venture in cooperation with the
RELM
Goals:
(1) To develop and test a range of viable earthquake-potential models
(not just one "consensus" model).
(2) To test these
models
againstalternative
existing &approaches
future geophysical
data.
thereby
including
such as the
more developmental physics-based models
(3) To examine the(establish
seismic hazard
implications
stationary
targets) of each model , which
will help …
• define existing uncertainties in seismic hazard analysis
• identify research topics needed to reduce these uncertainties
• identify which models are exportable to regions where options are fewer
RELM
Models Under Development
e.g., based on:
(1) geological fault data.
(2) historical seismicity.
(3) geodetic strain observations.
(4) stress interaction between earthquakes.
(5) spatial and temporal foreshock-aftershock statistics.
(6) Physical simulation models
(e.g., Rundle’s “Virtual California”)
Part II
SHA needs
Information Technology
SHA Must Interface Among:
Disciplines:
Scientists (of many ilks)
Engineers (research and practice)
Risk Analysts
Emergency Response Officials
The Public
Multiple Modeling and Database resources
(especially as we move toward physicsbased, system-level models).
Example from RELM
A wide variety of earthquake-forecast models
are under development.
How do we evaluate the hazard implications of
each?
Previous approach was to butcher existing SHA
Fortran code (custom hard-wire) to handle each
new model separately. This is no longer a
viable approach.
This led to the development of:
Java-based (object oriented) code for
Seismic Hazard Analysis
with a standard interface to allow any type of
earthquake-forecast model to plug in
this has subsequently evolved into a …
Framework for Object-Oriented
Seismic Hazard Analysis
(FOOSHA?)
GOAL: A general framework that will allow the
implementation of all foreseeable types of earthquakeforecast and ground motion models, as well as satisfy
the rapidly evolving analysis demands of the
engineering community and other user groups .
Our Developmental Approach
• Outline a proposed object-oriented SHA framework
(not yet
sureorthis
is a good
(conceptual
model
ontology)
&blueprint)
publish in SRL to
get feedback from the broader community.
• Begin implementing the core components in Java.
• Implement the framework in Pathway 1 of the SCEC
ITR Collaboration (expand ontology for KR&R,
utilize Grid computing and digital libraries).
• Implement the various RELM earthquake-forecast
models (many will be wrapped Fortran code).
• Modify framework as needed, hopefully leading to a
useful community standard.
Basic Framework (simplified).
Desired output is the probability that
something of concern will happen
over a specified time span
IMT
IML
Site
PQkn
Time Span
Earthquake
Forecast
Streaming
Potential
Earthquakes
(N total)
Intensity-Measure
Relationship
N
Prob( IML)  1 1 Prob( IML | PQkn ) * Prob( PQkn)
n1
Intensity-Measure Type/Level
a specification of what the
analyst (e.g., engineer) is
worried about
IMT
IML
Site
Time Span
Earthquake
Forecast
PQkn
Streaming
Potential
Earthquakes
(N total)
Intensity-Measure
Relationship
N
Prob( IML)  1 1 Prob( IML | PQkn ) * Prob( PQkn)
n1
Site and Potential Earthquake
The two main physical
objects used in the analysis
IMT
IML
Site
PQkn
Time Span
Earthquake
Forecast
Streaming
Potential
Earthquakes
(N total)
Intensity-Measure
Relationship
N
Prob( IML)  1 1 Prob( IML | PQkn ) * Prob( PQkn)
n1
Earthquake Forecast
Time Span
One of the two main model
components. A wide variety
of types will be developed.
IMT
IML
Site
PQkn
Earthquake
Forecast
Streaming
Potential
Earthquakes
(N total)
Intensity-Measure
Relationship
N
Prob( IML)  1 1 Prob( IML | PQkn ) * Prob( PQkn)
n1
Intensity-Measure Relationship
The other major model
component. A variety of these
will also be developed.
IMT
IML
Site
PQkn
Time Span
Earthquake
Forecast
Streaming
Potential
Earthquakes
(N total)
Intensity-Measure
Relationship
N
Prob( IML)  1 1 Prob( IML | PQkn ) * Prob( PQkn)
n1
IMT
IML
Site
PQkn
Intensity-Measure
Relationship
Intensity-Measure
Regressions
Prob(≥IML)
(attenuation relations)
SCEC ITR Collaboration:
Pathway 1
Pathway 2
Full Waveform
Modeling
Models will have to access shared data
resources (community databases)
For example …
Time
Span
PQk
List
Network
Earthquake
Catalog
Fault Activity
Database
Earthquake
Forecast
Historical
Earthquake
Catalog
GPS Data
(Velocity Vectors)
Community
Fault Model
Sharing data resources will help:
•
Avoid effort duplication
•
Help us determine where differences in model
predictions come from by ensuring that participants are
using consistent data constraints (a huge problem in
previous SHA efforts)
•
If data are machine readable from the host institution,
model predictions can easily be updated when
inevitable revisions or additions are made to the
database (previously data revisions have led to multiyear publication delays).
(all of these are critical to improving SHA)
Exactly how will the ITR
collaboration help?
(a few examples)
Grid Computing:
To enable run-time access to whatever high
performance computing resources are available at that
moment.
This will help reduce the time to generate a hazard
map, or a synthetic seismogram, from hours to
(hopefully) seconds.
Exactly how will the ITR
collaboration help?
(a few examples)
Grid Computing & Digital Libraries:
To enable interoperability among distributed
earthquake-forecast and ground-motion models and
the various databases they depend upon.
This will put maintenance responsibility solely on
the developer or host.
Exactly how will the ITR
collaboration help?
(a few examples)
Knowledge Representation and Reasoning (KR&R):
To keep track of the ontological attributes and
relationships that we don’t want to encumber the
Java code with.
e.g. Attributes of an Earthquake Forecast: the region and time-span it’s
applicable to; that it’s a monte-carlo simulation versus having adjustable
parameters; handling of epistemic versus aleatory uncertainties; whether
it’s a “next event” or “all event” prediction; …
Exactly how will the ITR
collaboration help?
(a few examples)
KR&R and Digital Libraries:
To enable smart eDatabase inquiries
(e.g., so that an appropriate probability model
can be constructed for a potential earthquake
based on what information is found in the
fault activity database).
Exactly how will the ITR
collaboration help?
(a few examples)
Digital Libraries:
To enable version tracking for purposes of SHA
reproducibility in an environment of continually
evolving models and databases.
Part III
Will EarthScope help SHA?
Will EarthScope help SHA?
Absolutely no question over the long term.
The most profound ways SHA will be
impacted by EarthScope are probably
unknown at this time.
EarthScope is a great idea.
Will EarthScope help SHA?
Big Concern:
In terms of SHA (at least) we’re not
making effective use of data that’s already
been collected.
e.g.,
Strong Motion Database (COSMOS)
(inadequate quality control)
Fault Activity Database
(lack thereof)
Will EarthScope help SHA?
The Problem is:
No one wants to pay for the grunt work needed to
put existing data into the most useful form.
Doing so would have a huge, direct, and
immediate impact on SHA.
Thus, from an SHA perspective, it’s hard for me
to rally around such a massive data gathering
effort as represented by EarthScope.
Will EarthScope help SHA?
If Earthscope gets funded:
I urge you to dedicate sufficient resourced for
making data available in a useful form.
I think 10% for IT is not enough.
Will EarthScope help SHA?
Biggest Concern:
If USArray ends up taking funds away
from ANSS, we will be in very serious
trouble with respect to improving SHA.
Final Questions
To what extent is EarthScope dedicated to
solving problems directly associated with
SHA?
How much is SHA being used to “sell”
EarthScope?