1. Earthquake Early Warning in California: Evaluating Hardware and Software
Margaret Hellweg, Richard Allen, Maren Böse,
Holly Brown, Georgia Cua, Doug Given,
Egill Hauksson, Tom Heaton, Tom Jordan,
Oleg Khainovski, Phil Maechling,
Doug Neuhauser, David Oppenheimer,
Kalpesh Solanki, Michael Zeleznik
University of California, Berkeley
California Institute of Technology
ETH Zurich
USGS Pasadena
Southern California Earthquake Center
USGS Menlo Park
Saya Systems
Funded by
CISN
California Integrated Seismic Network

2. What is earthquake early warning?
Seismic Stations
P-wave
fault
S-wave

3. What is earthquake early warning?
minimal telemetry delays
but less robust
warning time is the S-P time
best chance of warning at epicenter
Single station approach
1st Station to detect
P-wave arrival
Issue warning?
Seismic Stations
P-wave
fault
S-wave
CISN
California Integrated Seismic Network
EEW
 “On-site” method

4. What is earthquake early warning?
Multiple stations detect
P-wave arrival
Seismic Stations
P-wave
fault
S-wave

5. What is earthquake early warning?
Multiple stations detect
P-wave arrival
Estimate the extent of rupture
Seismic Stations
P-wave
fault
S-wave
Rupture

6. What is earthquake early warning?
Network based approach
Multiple stations detect
P-wave arrival
Estimate the extent of rupture
Issue warning
longer delay
but more robust
difficult to warn at epicenter
transmit warning ahead of shaking …greater warning times
Seismic Stations
P-wave
fault
CISN
California Integrated Seismic Network
EEW
S-wave
 “ElarmS” and
“Virtual Seismologist”
methods
Rupture
Shaking

7. What is earthquake early warning?
More stations detect
P-wave arrival
Estimate the extent of rupture
Issue warning
More information
 update warning
Seismic Stations
P-wave
fault
S-wave
Rupture
Shaking

8. California Integrated Seismic Network
August 2006 – July 2009
statewide testing
CISN
California Integrated Seismic Network
EEW
Goal: evaluate real-time performance of early warning algorithms
Aug ‘06
Year 1: Initial real-time outputs from EEW algorithms
Year 2: Statewide processing; Add uncertainty estimates; website display
Year 3: Evaluation of past and future performance
Year 1: Initial real-time outputs from EEW algorithms
Year 2: Statewide processing; Add uncertainty estimates; website display
Year 3: Evaluation of past and future performance
Currently at 28 mths
Jul ‘09
 specifications for implemented early warning system
equipment requirements (stations)
telemetry
processing
…the next generation of geophysical networks 8

9. California Integrated Seismic Network
August 2006 – July 2009
State-wide testing
CISN
California Integrated Seismic Network
EEW
Goal: evaluate real-time performance of early warning methods
Distributed waveform processing
Each algorithms runs at one datacenter
Reporting to SCEC
Three algorithms:
Archive/website
SCEC/USC
1. Onsite warning (S11A-1727)
Caltech/U. Taiwan
2. Virtual Seismologist (S11A-1724, S14B-06)
Caltech/ETH
3. ElarmS (S11A-1726 , S11A-1725)
UC Berkeley
Algorithm Evaluation:
UC Berkeley
USGS Menlo Park
Alert information and
summaries (S11A-1728)
SCEC/USC
Caltech/USGS Pasadena

10. τc-Pd On-site Warning Algorithm (S11A-1727)
CISN
California Integrated Seismic Network
EEW
τc-Pd On-site Warning Algorithm (S11A-1727)
: 58 earthquakes
with 3.0 ≤ Mw ≤ 5.4
Mw5.4 Chino Hills 07/29/2008:
60 triggers
Uncertainties after station corrections:
Mw: ±0.5; log(PGV): ±0.3; MMI: ±0.7
HH estimated Mw=4.4…6.5 (median: 5.6)
First estimate: Mw=5.6 (CI.PSR, 10 sec after O.T.) → 6 sec warning time at LA City Hall
Magnitude
Peak Ground Velocity
Warning delays
~ sec 10

11. Virtual seismologist (S11A-1724, S14B-06)
CISN
California Integrated Seismic Network
EEW
Virtual seismologist (S11A-1724, S14B-06)
Regional network approach
Bayesian approach: Include likelihood of earthquake given past observations
e.g. Gutenberg-Richter relation and distribution of previous seismicity
Triggers on P-waves
Uses envelope functions of waveforms
Magnitude derived from 3 sec of the P-wave
Predicts the distribution of ground shaking
Finite fault discriminate
Cua & Heaton 2007
Real-time in SC since 13 July 2008
Initial estimate within s
of origin time
Mw: ±1 for 97% of events
Fig captions are included (cropped off display)
Left: Figure 5.11 from Yamada’s thesis: probability that a station is near source based on their discriminat function
Right: Figs 6.5 and 6.7 from Yamada’s thesis: back projection of station displacement onto the fault; backprojected slip estimates and the estimated slip from kinematic inversions
Next steps:
Statewide processing
Site conditions
EQ/noise discrimination
Yamada & Heaton in review

12. California Integrated Seismic Network
CISN
California Integrated Seismic Network
EEW
ElarmS-RT (S11A-1726 , S11A-1725)
Statewide (Nov 08) (waveform processing at UCB, USGS-MP and Caltech)
Regional network approach
Triggers on P-waves, uses arrival times, frequency and amplitude of P-wave. Uses PGA and PGV as they become available.
Predicts the distribution of ground shaking using ShakeMap approach  AlertMaps
AlertMaps – ElarmS-RT – MW5.4 October 30, 2007
trigger time
+ 1 sec
+ 2 sec
+ 3 sec

13. Testing Algorithm Performance (S11A-1728)
CISN
California Integrated Seismic Network
EEW
Testing Algorithm Performance (S11A-1728)
Performance summaries (Location/Magnitude currently implemented) generated every night (2mo,4mo,6mo and cumulative time periods).
Generating Summaries for ElarmS (UCB), TauC (Caltech),and Virtual Seismologist ( ETH)
CISN EEW Testing site available at: Login: guest Pwd: cisneew
Magnitude Accuracy
Warning Delay with current CISN network
Theoretical Warning Delay with no telemetry or processing delays 13

14. California Integrated Seismic Network
CISN
California Integrated Seismic Network
EEW
Perspectives …