Forecasting earthquakes: the state of the art Andrew Bell and Ian Main

Forecasting earthquakes:
the state of the art
Andrew Bell and Ian Main
School of GeoSciences,
University of Edinburgh
Uniform California Earthquake Rupture Forecast (2007)
L’Aquila April 2009
Official Italian seismic classification map (2006)
Earthquake forecasting
• Earthquake prediction:
“the prediction of the location, time and magnitude of
an earthquake, in advance, within narrow limits, above
chance” Richter (1954)
• Types of forecast:
– Time-independent hazard
– Time-dependent hazard
– Earthquake forecasting
• Forecast requirements:
– Data > Models > Tests
• Full quantification of
Time-independent hazard
• Data:
– Earthquake history, fault maps,
long-term deformation rates etc.
• Model:
– Random (Poisson) process with time
– Local smoothing parameters
– Path & site effects
• Output:
– Probability of exceeding certain
ground shaking in X years
ESC-SESAME map (2003)
Time-dependent hazard
• Includes non-random recurrence
time information
Shimazaki & Nakata (1980)
Uniform California Earthquake Rupture Forecast (2007)
Statistical earthquake forecasting models
• Epidemic-Type Aftershock Model
• Short-Term Earthquake Probability
California 24-Hour Aftershock Forecast Map
Physical earthquake forecasting models
• Quantifiable physical perturbation
• Quantifiable Earth response
• e.g. Coulomb stress transfer +
rate & state friction
Coulomb stress changes fro Landers & Big Bear
earthquakes,1992, King et al. (1994)
Testing earthquake forecasts
• Testing Centers
RELM tests, SRL 2007
ERI, Japan
ETH, Switzerland
GNS, New Zealand
SCEC, United States
• Testing Regions
Northwest Pacific
Southwest Pacific
New Zealand
Coming soon…
• Improved data quality/quantity
• Model development
– Composite physical + statistical models
– Scenario modelling
– Improved uncertainty quantification
• Advanced testing protocols
– Global-scale model testing
– Quantification of forecast consistency,
quality (reliability, accuracy etc.) and
value (Murphy, 1993)
European VEBSN stations, NERIES
Is any of this practicable for DRR?
• Significant probability gains, but low absolute probabilities
• How can potentially small changes in probability be used?
Cost-benefit balance for actions based on potentially small probability gains?
L’Aquila, 2009. Photo: Daily Telegraph