Real-Time Estimation of Earthquake Location and Magnitude for Seismic Early Warning in Campania Region, southern Italy A.
Download ReportTranscript Real-Time Estimation of Earthquake Location and Magnitude for Seismic Early Warning in Campania Region, southern Italy A.
Slide 1
Real-Time Estimation of Earthquake Location
and Magnitude for Seismic Early Warning in
Campania Region, southern Italy
A. Zollo and RISSC-Lab Research Group*
with A.Lomax (A.Lomax Scientific Software)
*
is a joint seismological research group between University of
Naples - Dept of Physics and INGV – Osservatorio Vesuviano
Work Motivation
• Development and testing of a Seismic EarlyWarning System for automated risk mitigation
actions in Campania Region
• Need for robust and reliable real-time estimates
of eqk location and magnitude to be obtained in
an evolving, continually updated form.
• Need to provide with parameter uncertainty
variation with time engineering structural
control
Historical Earthquakes
1980 Irpinia
earthquake,
Ms=6.9
Recent Seismicity
Early Warning
Network
INGV catalogue
(1981-2002)
M2.5
29 sites
Osiris 24-bit Data Logger
6 channels:
3 accelerometers
3 seismometers
(Short Period or
Broad Band)
Real time data analysis
SEW System Peculiarities
Characteristic times:
3-5 s
1.5 - 3.5 s
16 - 18 s
eqk at 4-16 km
depth
Latency
&
computation
60 km
22 - 24 s
80 km
28 - 30 s
100 km
time
To
TP first
TS target
High spatial density :
Station spacing < 15 km
Wide-Dynamics:
Unsaturated signals up to 1 g
Real-Time Earthquake Location
Basic Ideas:
Constraint from “not-yet-triggered” stations
Tracing and crossing Equal Differential
Time (EDT) surfaces
Probabilistic estimation of eqk location vs
time (Evolutionary Approach)
Real-time Evolutionary Location
• When a first station Sn triggers at tn = tnow,
we can already place limits on a pdf
volume that is likely to contain the
hypocenter (Voronoi cell). These limits
are given by conditional EDT surfaces on
which the P travel time to the first
triggering station A is equal to the traveltime to each of the not-yet-triggered
stations.
• As the current time tnow progresses we
gain the additional information that the
not-yet-triggered stations can only trigger
with tl > tnow
• When the second and later stations
trigger, we construct standard, true EDT
surfaces between each pair of the
triggered stations.These EDT surfaces
are stacked with the volume defined by
the not-yet-triggered stations to form the
current hypocentral pdf volume.
Synthetic Examples
Earthquake location probability
Seconds from first trigger
Seconds from earthquake Origin Time
Triggered stations
Earthquake location probability
Real-Time Magnitude Estimate
Basic Ideas
• Use both early P- and S-wave information based
on the high density / wide dynamics of the
network
• Correlate low-pass filtered peak amplitudes with
Magnitude in increasing time windows
• Regression analysis based on the European
strong motion Data-Base (ESD, Ambraseys,
2004)
European Strong Motion Data Base*
* Ambraseys et al. (2004)
• 207 Events with 4≤MW≤7.4 (Kokaeli,1999)
• 376 three-component records
• Epicentral distance ≤ 50 km
• Low-pass filter: 3 Hz
• Magnitude bin: 0.3
Vertical
Measurement of Peak Amplitude
2-sec
H(t)=NS2(t)+EW2(t)
Horizontal
1-sec
3Hz low-pass filtered acceleration at station Bagnoli
(1980, Irpinia Eqk, Ms=6.9) – Epicentral Distance: 20 Km
Log(displacement)
Log(PGDt) vs Magnitude
P-wave
S-wave
Mean value
2-Weighted
Standard Error
Single data point
magnitude
A possible explanation
The “far-field” approximation for displacement (f=3Hz, D> 5-6 km):
moment rate
Slip-rate vs M
Active slip area vs M
Rupture kinematics
Rupture dynamics
Dynamic stress drop vs M
Rise-time vs M
The observed correlation between log(PGXt) and Magnitude would imply that
“dynamic stress release” and/or “rise-time”
scale with earthquake size in the very early stage of seismic ruptures
Summary
A high-density, wide-dynamics seismic network is being
installed in southern Italy for “regional” early-warning
applications
The system will implement a real-time eqk location method
based on an evolutionary, probabilistic approach
Early P- and S- signal amplitudes (less the 2 sec from first
arrival) correlate with magnitude (4≤Mw≤7.4) as from the
analysis of the European Strong Motion Data Base
A combination of magnitude estimations obtained by “early
P/S peak amplitudes” and “predominant periods” (Allen &
Kanamori,2003) measured at different stations as a
function of time may significantly improve the accuracy of
the earthquake size estimation in real-time procedures.
The End
Multiple Events
Each time a new pick is available,
the algorithm:
1. Temporarily associates the pick
to the current event
2. Relocates the event
3. Checks the travel-time RMS for
the maximum likelihood
hypocenter
4. If RMS < RMSthresh the pick is
definitively associated,
otherwise a new event is
declared
Vertical
Modulus of
horizontal
components
Strong Motion Data
De Natale et al., BSSA,1987
Kanamori & Rivera, BSSA, 2004
acceleration
velocity
displacement
Log(PGXt) vs Magnitude
Early P-Wave
Early S-Wave
Slide 2
Real-Time Estimation of Earthquake Location
and Magnitude for Seismic Early Warning in
Campania Region, southern Italy
A. Zollo and RISSC-Lab Research Group*
with A.Lomax (A.Lomax Scientific Software)
*
is a joint seismological research group between University of
Naples - Dept of Physics and INGV – Osservatorio Vesuviano
Work Motivation
• Development and testing of a Seismic EarlyWarning System for automated risk mitigation
actions in Campania Region
• Need for robust and reliable real-time estimates
of eqk location and magnitude to be obtained in
an evolving, continually updated form.
• Need to provide with parameter uncertainty
variation with time engineering structural
control
Historical Earthquakes
1980 Irpinia
earthquake,
Ms=6.9
Recent Seismicity
Early Warning
Network
INGV catalogue
(1981-2002)
M2.5
29 sites
Osiris 24-bit Data Logger
6 channels:
3 accelerometers
3 seismometers
(Short Period or
Broad Band)
Real time data analysis
SEW System Peculiarities
Characteristic times:
3-5 s
1.5 - 3.5 s
16 - 18 s
eqk at 4-16 km
depth
Latency
&
computation
60 km
22 - 24 s
80 km
28 - 30 s
100 km
time
To
TP first
TS target
High spatial density :
Station spacing < 15 km
Wide-Dynamics:
Unsaturated signals up to 1 g
Real-Time Earthquake Location
Basic Ideas:
Constraint from “not-yet-triggered” stations
Tracing and crossing Equal Differential
Time (EDT) surfaces
Probabilistic estimation of eqk location vs
time (Evolutionary Approach)
Real-time Evolutionary Location
• When a first station Sn triggers at tn = tnow,
we can already place limits on a pdf
volume that is likely to contain the
hypocenter (Voronoi cell). These limits
are given by conditional EDT surfaces on
which the P travel time to the first
triggering station A is equal to the traveltime to each of the not-yet-triggered
stations.
• As the current time tnow progresses we
gain the additional information that the
not-yet-triggered stations can only trigger
with tl > tnow
• When the second and later stations
trigger, we construct standard, true EDT
surfaces between each pair of the
triggered stations.These EDT surfaces
are stacked with the volume defined by
the not-yet-triggered stations to form the
current hypocentral pdf volume.
Synthetic Examples
Earthquake location probability
Seconds from first trigger
Seconds from earthquake Origin Time
Triggered stations
Earthquake location probability
Real-Time Magnitude Estimate
Basic Ideas
• Use both early P- and S-wave information based
on the high density / wide dynamics of the
network
• Correlate low-pass filtered peak amplitudes with
Magnitude in increasing time windows
• Regression analysis based on the European
strong motion Data-Base (ESD, Ambraseys,
2004)
European Strong Motion Data Base*
* Ambraseys et al. (2004)
• 207 Events with 4≤MW≤7.4 (Kokaeli,1999)
• 376 three-component records
• Epicentral distance ≤ 50 km
• Low-pass filter: 3 Hz
• Magnitude bin: 0.3
Vertical
Measurement of Peak Amplitude
2-sec
H(t)=NS2(t)+EW2(t)
Horizontal
1-sec
3Hz low-pass filtered acceleration at station Bagnoli
(1980, Irpinia Eqk, Ms=6.9) – Epicentral Distance: 20 Km
Log(displacement)
Log(PGDt) vs Magnitude
P-wave
S-wave
Mean value
2-Weighted
Standard Error
Single data point
magnitude
A possible explanation
The “far-field” approximation for displacement (f=3Hz, D> 5-6 km):
moment rate
Slip-rate vs M
Active slip area vs M
Rupture kinematics
Rupture dynamics
Dynamic stress drop vs M
Rise-time vs M
The observed correlation between log(PGXt) and Magnitude would imply that
“dynamic stress release” and/or “rise-time”
scale with earthquake size in the very early stage of seismic ruptures
Summary
A high-density, wide-dynamics seismic network is being
installed in southern Italy for “regional” early-warning
applications
The system will implement a real-time eqk location method
based on an evolutionary, probabilistic approach
Early P- and S- signal amplitudes (less the 2 sec from first
arrival) correlate with magnitude (4≤Mw≤7.4) as from the
analysis of the European Strong Motion Data Base
A combination of magnitude estimations obtained by “early
P/S peak amplitudes” and “predominant periods” (Allen &
Kanamori,2003) measured at different stations as a
function of time may significantly improve the accuracy of
the earthquake size estimation in real-time procedures.
The End
Multiple Events
Each time a new pick is available,
the algorithm:
1. Temporarily associates the pick
to the current event
2. Relocates the event
3. Checks the travel-time RMS for
the maximum likelihood
hypocenter
4. If RMS < RMSthresh the pick is
definitively associated,
otherwise a new event is
declared
Vertical
Modulus of
horizontal
components
Strong Motion Data
De Natale et al., BSSA,1987
Kanamori & Rivera, BSSA, 2004
acceleration
velocity
displacement
Log(PGXt) vs Magnitude
Early P-Wave
Early S-Wave
Slide 3
Real-Time Estimation of Earthquake Location
and Magnitude for Seismic Early Warning in
Campania Region, southern Italy
A. Zollo and RISSC-Lab Research Group*
with A.Lomax (A.Lomax Scientific Software)
*
is a joint seismological research group between University of
Naples - Dept of Physics and INGV – Osservatorio Vesuviano
Work Motivation
• Development and testing of a Seismic EarlyWarning System for automated risk mitigation
actions in Campania Region
• Need for robust and reliable real-time estimates
of eqk location and magnitude to be obtained in
an evolving, continually updated form.
• Need to provide with parameter uncertainty
variation with time engineering structural
control
Historical Earthquakes
1980 Irpinia
earthquake,
Ms=6.9
Recent Seismicity
Early Warning
Network
INGV catalogue
(1981-2002)
M2.5
29 sites
Osiris 24-bit Data Logger
6 channels:
3 accelerometers
3 seismometers
(Short Period or
Broad Band)
Real time data analysis
SEW System Peculiarities
Characteristic times:
3-5 s
1.5 - 3.5 s
16 - 18 s
eqk at 4-16 km
depth
Latency
&
computation
60 km
22 - 24 s
80 km
28 - 30 s
100 km
time
To
TP first
TS target
High spatial density :
Station spacing < 15 km
Wide-Dynamics:
Unsaturated signals up to 1 g
Real-Time Earthquake Location
Basic Ideas:
Constraint from “not-yet-triggered” stations
Tracing and crossing Equal Differential
Time (EDT) surfaces
Probabilistic estimation of eqk location vs
time (Evolutionary Approach)
Real-time Evolutionary Location
• When a first station Sn triggers at tn = tnow,
we can already place limits on a pdf
volume that is likely to contain the
hypocenter (Voronoi cell). These limits
are given by conditional EDT surfaces on
which the P travel time to the first
triggering station A is equal to the traveltime to each of the not-yet-triggered
stations.
• As the current time tnow progresses we
gain the additional information that the
not-yet-triggered stations can only trigger
with tl > tnow
• When the second and later stations
trigger, we construct standard, true EDT
surfaces between each pair of the
triggered stations.These EDT surfaces
are stacked with the volume defined by
the not-yet-triggered stations to form the
current hypocentral pdf volume.
Synthetic Examples
Earthquake location probability
Seconds from first trigger
Seconds from earthquake Origin Time
Triggered stations
Earthquake location probability
Real-Time Magnitude Estimate
Basic Ideas
• Use both early P- and S-wave information based
on the high density / wide dynamics of the
network
• Correlate low-pass filtered peak amplitudes with
Magnitude in increasing time windows
• Regression analysis based on the European
strong motion Data-Base (ESD, Ambraseys,
2004)
European Strong Motion Data Base*
* Ambraseys et al. (2004)
• 207 Events with 4≤MW≤7.4 (Kokaeli,1999)
• 376 three-component records
• Epicentral distance ≤ 50 km
• Low-pass filter: 3 Hz
• Magnitude bin: 0.3
Vertical
Measurement of Peak Amplitude
2-sec
H(t)=NS2(t)+EW2(t)
Horizontal
1-sec
3Hz low-pass filtered acceleration at station Bagnoli
(1980, Irpinia Eqk, Ms=6.9) – Epicentral Distance: 20 Km
Log(displacement)
Log(PGDt) vs Magnitude
P-wave
S-wave
Mean value
2-Weighted
Standard Error
Single data point
magnitude
A possible explanation
The “far-field” approximation for displacement (f=3Hz, D> 5-6 km):
moment rate
Slip-rate vs M
Active slip area vs M
Rupture kinematics
Rupture dynamics
Dynamic stress drop vs M
Rise-time vs M
The observed correlation between log(PGXt) and Magnitude would imply that
“dynamic stress release” and/or “rise-time”
scale with earthquake size in the very early stage of seismic ruptures
Summary
A high-density, wide-dynamics seismic network is being
installed in southern Italy for “regional” early-warning
applications
The system will implement a real-time eqk location method
based on an evolutionary, probabilistic approach
Early P- and S- signal amplitudes (less the 2 sec from first
arrival) correlate with magnitude (4≤Mw≤7.4) as from the
analysis of the European Strong Motion Data Base
A combination of magnitude estimations obtained by “early
P/S peak amplitudes” and “predominant periods” (Allen &
Kanamori,2003) measured at different stations as a
function of time may significantly improve the accuracy of
the earthquake size estimation in real-time procedures.
The End
Multiple Events
Each time a new pick is available,
the algorithm:
1. Temporarily associates the pick
to the current event
2. Relocates the event
3. Checks the travel-time RMS for
the maximum likelihood
hypocenter
4. If RMS < RMSthresh the pick is
definitively associated,
otherwise a new event is
declared
Vertical
Modulus of
horizontal
components
Strong Motion Data
De Natale et al., BSSA,1987
Kanamori & Rivera, BSSA, 2004
acceleration
velocity
displacement
Log(PGXt) vs Magnitude
Early P-Wave
Early S-Wave
Slide 4
Real-Time Estimation of Earthquake Location
and Magnitude for Seismic Early Warning in
Campania Region, southern Italy
A. Zollo and RISSC-Lab Research Group*
with A.Lomax (A.Lomax Scientific Software)
*
is a joint seismological research group between University of
Naples - Dept of Physics and INGV – Osservatorio Vesuviano
Work Motivation
• Development and testing of a Seismic EarlyWarning System for automated risk mitigation
actions in Campania Region
• Need for robust and reliable real-time estimates
of eqk location and magnitude to be obtained in
an evolving, continually updated form.
• Need to provide with parameter uncertainty
variation with time engineering structural
control
Historical Earthquakes
1980 Irpinia
earthquake,
Ms=6.9
Recent Seismicity
Early Warning
Network
INGV catalogue
(1981-2002)
M2.5
29 sites
Osiris 24-bit Data Logger
6 channels:
3 accelerometers
3 seismometers
(Short Period or
Broad Band)
Real time data analysis
SEW System Peculiarities
Characteristic times:
3-5 s
1.5 - 3.5 s
16 - 18 s
eqk at 4-16 km
depth
Latency
&
computation
60 km
22 - 24 s
80 km
28 - 30 s
100 km
time
To
TP first
TS target
High spatial density :
Station spacing < 15 km
Wide-Dynamics:
Unsaturated signals up to 1 g
Real-Time Earthquake Location
Basic Ideas:
Constraint from “not-yet-triggered” stations
Tracing and crossing Equal Differential
Time (EDT) surfaces
Probabilistic estimation of eqk location vs
time (Evolutionary Approach)
Real-time Evolutionary Location
• When a first station Sn triggers at tn = tnow,
we can already place limits on a pdf
volume that is likely to contain the
hypocenter (Voronoi cell). These limits
are given by conditional EDT surfaces on
which the P travel time to the first
triggering station A is equal to the traveltime to each of the not-yet-triggered
stations.
• As the current time tnow progresses we
gain the additional information that the
not-yet-triggered stations can only trigger
with tl > tnow
• When the second and later stations
trigger, we construct standard, true EDT
surfaces between each pair of the
triggered stations.These EDT surfaces
are stacked with the volume defined by
the not-yet-triggered stations to form the
current hypocentral pdf volume.
Synthetic Examples
Earthquake location probability
Seconds from first trigger
Seconds from earthquake Origin Time
Triggered stations
Earthquake location probability
Real-Time Magnitude Estimate
Basic Ideas
• Use both early P- and S-wave information based
on the high density / wide dynamics of the
network
• Correlate low-pass filtered peak amplitudes with
Magnitude in increasing time windows
• Regression analysis based on the European
strong motion Data-Base (ESD, Ambraseys,
2004)
European Strong Motion Data Base*
* Ambraseys et al. (2004)
• 207 Events with 4≤MW≤7.4 (Kokaeli,1999)
• 376 three-component records
• Epicentral distance ≤ 50 km
• Low-pass filter: 3 Hz
• Magnitude bin: 0.3
Vertical
Measurement of Peak Amplitude
2-sec
H(t)=NS2(t)+EW2(t)
Horizontal
1-sec
3Hz low-pass filtered acceleration at station Bagnoli
(1980, Irpinia Eqk, Ms=6.9) – Epicentral Distance: 20 Km
Log(displacement)
Log(PGDt) vs Magnitude
P-wave
S-wave
Mean value
2-Weighted
Standard Error
Single data point
magnitude
A possible explanation
The “far-field” approximation for displacement (f=3Hz, D> 5-6 km):
moment rate
Slip-rate vs M
Active slip area vs M
Rupture kinematics
Rupture dynamics
Dynamic stress drop vs M
Rise-time vs M
The observed correlation between log(PGXt) and Magnitude would imply that
“dynamic stress release” and/or “rise-time”
scale with earthquake size in the very early stage of seismic ruptures
Summary
A high-density, wide-dynamics seismic network is being
installed in southern Italy for “regional” early-warning
applications
The system will implement a real-time eqk location method
based on an evolutionary, probabilistic approach
Early P- and S- signal amplitudes (less the 2 sec from first
arrival) correlate with magnitude (4≤Mw≤7.4) as from the
analysis of the European Strong Motion Data Base
A combination of magnitude estimations obtained by “early
P/S peak amplitudes” and “predominant periods” (Allen &
Kanamori,2003) measured at different stations as a
function of time may significantly improve the accuracy of
the earthquake size estimation in real-time procedures.
The End
Multiple Events
Each time a new pick is available,
the algorithm:
1. Temporarily associates the pick
to the current event
2. Relocates the event
3. Checks the travel-time RMS for
the maximum likelihood
hypocenter
4. If RMS < RMSthresh the pick is
definitively associated,
otherwise a new event is
declared
Vertical
Modulus of
horizontal
components
Strong Motion Data
De Natale et al., BSSA,1987
Kanamori & Rivera, BSSA, 2004
acceleration
velocity
displacement
Log(PGXt) vs Magnitude
Early P-Wave
Early S-Wave
Slide 5
Real-Time Estimation of Earthquake Location
and Magnitude for Seismic Early Warning in
Campania Region, southern Italy
A. Zollo and RISSC-Lab Research Group*
with A.Lomax (A.Lomax Scientific Software)
*
is a joint seismological research group between University of
Naples - Dept of Physics and INGV – Osservatorio Vesuviano
Work Motivation
• Development and testing of a Seismic EarlyWarning System for automated risk mitigation
actions in Campania Region
• Need for robust and reliable real-time estimates
of eqk location and magnitude to be obtained in
an evolving, continually updated form.
• Need to provide with parameter uncertainty
variation with time engineering structural
control
Historical Earthquakes
1980 Irpinia
earthquake,
Ms=6.9
Recent Seismicity
Early Warning
Network
INGV catalogue
(1981-2002)
M2.5
29 sites
Osiris 24-bit Data Logger
6 channels:
3 accelerometers
3 seismometers
(Short Period or
Broad Band)
Real time data analysis
SEW System Peculiarities
Characteristic times:
3-5 s
1.5 - 3.5 s
16 - 18 s
eqk at 4-16 km
depth
Latency
&
computation
60 km
22 - 24 s
80 km
28 - 30 s
100 km
time
To
TP first
TS target
High spatial density :
Station spacing < 15 km
Wide-Dynamics:
Unsaturated signals up to 1 g
Real-Time Earthquake Location
Basic Ideas:
Constraint from “not-yet-triggered” stations
Tracing and crossing Equal Differential
Time (EDT) surfaces
Probabilistic estimation of eqk location vs
time (Evolutionary Approach)
Real-time Evolutionary Location
• When a first station Sn triggers at tn = tnow,
we can already place limits on a pdf
volume that is likely to contain the
hypocenter (Voronoi cell). These limits
are given by conditional EDT surfaces on
which the P travel time to the first
triggering station A is equal to the traveltime to each of the not-yet-triggered
stations.
• As the current time tnow progresses we
gain the additional information that the
not-yet-triggered stations can only trigger
with tl > tnow
• When the second and later stations
trigger, we construct standard, true EDT
surfaces between each pair of the
triggered stations.These EDT surfaces
are stacked with the volume defined by
the not-yet-triggered stations to form the
current hypocentral pdf volume.
Synthetic Examples
Earthquake location probability
Seconds from first trigger
Seconds from earthquake Origin Time
Triggered stations
Earthquake location probability
Real-Time Magnitude Estimate
Basic Ideas
• Use both early P- and S-wave information based
on the high density / wide dynamics of the
network
• Correlate low-pass filtered peak amplitudes with
Magnitude in increasing time windows
• Regression analysis based on the European
strong motion Data-Base (ESD, Ambraseys,
2004)
European Strong Motion Data Base*
* Ambraseys et al. (2004)
• 207 Events with 4≤MW≤7.4 (Kokaeli,1999)
• 376 three-component records
• Epicentral distance ≤ 50 km
• Low-pass filter: 3 Hz
• Magnitude bin: 0.3
Vertical
Measurement of Peak Amplitude
2-sec
H(t)=NS2(t)+EW2(t)
Horizontal
1-sec
3Hz low-pass filtered acceleration at station Bagnoli
(1980, Irpinia Eqk, Ms=6.9) – Epicentral Distance: 20 Km
Log(displacement)
Log(PGDt) vs Magnitude
P-wave
S-wave
Mean value
2-Weighted
Standard Error
Single data point
magnitude
A possible explanation
The “far-field” approximation for displacement (f=3Hz, D> 5-6 km):
moment rate
Slip-rate vs M
Active slip area vs M
Rupture kinematics
Rupture dynamics
Dynamic stress drop vs M
Rise-time vs M
The observed correlation between log(PGXt) and Magnitude would imply that
“dynamic stress release” and/or “rise-time”
scale with earthquake size in the very early stage of seismic ruptures
Summary
A high-density, wide-dynamics seismic network is being
installed in southern Italy for “regional” early-warning
applications
The system will implement a real-time eqk location method
based on an evolutionary, probabilistic approach
Early P- and S- signal amplitudes (less the 2 sec from first
arrival) correlate with magnitude (4≤Mw≤7.4) as from the
analysis of the European Strong Motion Data Base
A combination of magnitude estimations obtained by “early
P/S peak amplitudes” and “predominant periods” (Allen &
Kanamori,2003) measured at different stations as a
function of time may significantly improve the accuracy of
the earthquake size estimation in real-time procedures.
The End
Multiple Events
Each time a new pick is available,
the algorithm:
1. Temporarily associates the pick
to the current event
2. Relocates the event
3. Checks the travel-time RMS for
the maximum likelihood
hypocenter
4. If RMS < RMSthresh the pick is
definitively associated,
otherwise a new event is
declared
Vertical
Modulus of
horizontal
components
Strong Motion Data
De Natale et al., BSSA,1987
Kanamori & Rivera, BSSA, 2004
acceleration
velocity
displacement
Log(PGXt) vs Magnitude
Early P-Wave
Early S-Wave
Slide 6
Real-Time Estimation of Earthquake Location
and Magnitude for Seismic Early Warning in
Campania Region, southern Italy
A. Zollo and RISSC-Lab Research Group*
with A.Lomax (A.Lomax Scientific Software)
*
is a joint seismological research group between University of
Naples - Dept of Physics and INGV – Osservatorio Vesuviano
Work Motivation
• Development and testing of a Seismic EarlyWarning System for automated risk mitigation
actions in Campania Region
• Need for robust and reliable real-time estimates
of eqk location and magnitude to be obtained in
an evolving, continually updated form.
• Need to provide with parameter uncertainty
variation with time engineering structural
control
Historical Earthquakes
1980 Irpinia
earthquake,
Ms=6.9
Recent Seismicity
Early Warning
Network
INGV catalogue
(1981-2002)
M2.5
29 sites
Osiris 24-bit Data Logger
6 channels:
3 accelerometers
3 seismometers
(Short Period or
Broad Band)
Real time data analysis
SEW System Peculiarities
Characteristic times:
3-5 s
1.5 - 3.5 s
16 - 18 s
eqk at 4-16 km
depth
Latency
&
computation
60 km
22 - 24 s
80 km
28 - 30 s
100 km
time
To
TP first
TS target
High spatial density :
Station spacing < 15 km
Wide-Dynamics:
Unsaturated signals up to 1 g
Real-Time Earthquake Location
Basic Ideas:
Constraint from “not-yet-triggered” stations
Tracing and crossing Equal Differential
Time (EDT) surfaces
Probabilistic estimation of eqk location vs
time (Evolutionary Approach)
Real-time Evolutionary Location
• When a first station Sn triggers at tn = tnow,
we can already place limits on a pdf
volume that is likely to contain the
hypocenter (Voronoi cell). These limits
are given by conditional EDT surfaces on
which the P travel time to the first
triggering station A is equal to the traveltime to each of the not-yet-triggered
stations.
• As the current time tnow progresses we
gain the additional information that the
not-yet-triggered stations can only trigger
with tl > tnow
• When the second and later stations
trigger, we construct standard, true EDT
surfaces between each pair of the
triggered stations.These EDT surfaces
are stacked with the volume defined by
the not-yet-triggered stations to form the
current hypocentral pdf volume.
Synthetic Examples
Earthquake location probability
Seconds from first trigger
Seconds from earthquake Origin Time
Triggered stations
Earthquake location probability
Real-Time Magnitude Estimate
Basic Ideas
• Use both early P- and S-wave information based
on the high density / wide dynamics of the
network
• Correlate low-pass filtered peak amplitudes with
Magnitude in increasing time windows
• Regression analysis based on the European
strong motion Data-Base (ESD, Ambraseys,
2004)
European Strong Motion Data Base*
* Ambraseys et al. (2004)
• 207 Events with 4≤MW≤7.4 (Kokaeli,1999)
• 376 three-component records
• Epicentral distance ≤ 50 km
• Low-pass filter: 3 Hz
• Magnitude bin: 0.3
Vertical
Measurement of Peak Amplitude
2-sec
H(t)=NS2(t)+EW2(t)
Horizontal
1-sec
3Hz low-pass filtered acceleration at station Bagnoli
(1980, Irpinia Eqk, Ms=6.9) – Epicentral Distance: 20 Km
Log(displacement)
Log(PGDt) vs Magnitude
P-wave
S-wave
Mean value
2-Weighted
Standard Error
Single data point
magnitude
A possible explanation
The “far-field” approximation for displacement (f=3Hz, D> 5-6 km):
moment rate
Slip-rate vs M
Active slip area vs M
Rupture kinematics
Rupture dynamics
Dynamic stress drop vs M
Rise-time vs M
The observed correlation between log(PGXt) and Magnitude would imply that
“dynamic stress release” and/or “rise-time”
scale with earthquake size in the very early stage of seismic ruptures
Summary
A high-density, wide-dynamics seismic network is being
installed in southern Italy for “regional” early-warning
applications
The system will implement a real-time eqk location method
based on an evolutionary, probabilistic approach
Early P- and S- signal amplitudes (less the 2 sec from first
arrival) correlate with magnitude (4≤Mw≤7.4) as from the
analysis of the European Strong Motion Data Base
A combination of magnitude estimations obtained by “early
P/S peak amplitudes” and “predominant periods” (Allen &
Kanamori,2003) measured at different stations as a
function of time may significantly improve the accuracy of
the earthquake size estimation in real-time procedures.
The End
Multiple Events
Each time a new pick is available,
the algorithm:
1. Temporarily associates the pick
to the current event
2. Relocates the event
3. Checks the travel-time RMS for
the maximum likelihood
hypocenter
4. If RMS < RMSthresh the pick is
definitively associated,
otherwise a new event is
declared
Vertical
Modulus of
horizontal
components
Strong Motion Data
De Natale et al., BSSA,1987
Kanamori & Rivera, BSSA, 2004
acceleration
velocity
displacement
Log(PGXt) vs Magnitude
Early P-Wave
Early S-Wave
Slide 7
Real-Time Estimation of Earthquake Location
and Magnitude for Seismic Early Warning in
Campania Region, southern Italy
A. Zollo and RISSC-Lab Research Group*
with A.Lomax (A.Lomax Scientific Software)
*
is a joint seismological research group between University of
Naples - Dept of Physics and INGV – Osservatorio Vesuviano
Work Motivation
• Development and testing of a Seismic EarlyWarning System for automated risk mitigation
actions in Campania Region
• Need for robust and reliable real-time estimates
of eqk location and magnitude to be obtained in
an evolving, continually updated form.
• Need to provide with parameter uncertainty
variation with time engineering structural
control
Historical Earthquakes
1980 Irpinia
earthquake,
Ms=6.9
Recent Seismicity
Early Warning
Network
INGV catalogue
(1981-2002)
M2.5
29 sites
Osiris 24-bit Data Logger
6 channels:
3 accelerometers
3 seismometers
(Short Period or
Broad Band)
Real time data analysis
SEW System Peculiarities
Characteristic times:
3-5 s
1.5 - 3.5 s
16 - 18 s
eqk at 4-16 km
depth
Latency
&
computation
60 km
22 - 24 s
80 km
28 - 30 s
100 km
time
To
TP first
TS target
High spatial density :
Station spacing < 15 km
Wide-Dynamics:
Unsaturated signals up to 1 g
Real-Time Earthquake Location
Basic Ideas:
Constraint from “not-yet-triggered” stations
Tracing and crossing Equal Differential
Time (EDT) surfaces
Probabilistic estimation of eqk location vs
time (Evolutionary Approach)
Real-time Evolutionary Location
• When a first station Sn triggers at tn = tnow,
we can already place limits on a pdf
volume that is likely to contain the
hypocenter (Voronoi cell). These limits
are given by conditional EDT surfaces on
which the P travel time to the first
triggering station A is equal to the traveltime to each of the not-yet-triggered
stations.
• As the current time tnow progresses we
gain the additional information that the
not-yet-triggered stations can only trigger
with tl > tnow
• When the second and later stations
trigger, we construct standard, true EDT
surfaces between each pair of the
triggered stations.These EDT surfaces
are stacked with the volume defined by
the not-yet-triggered stations to form the
current hypocentral pdf volume.
Synthetic Examples
Earthquake location probability
Seconds from first trigger
Seconds from earthquake Origin Time
Triggered stations
Earthquake location probability
Real-Time Magnitude Estimate
Basic Ideas
• Use both early P- and S-wave information based
on the high density / wide dynamics of the
network
• Correlate low-pass filtered peak amplitudes with
Magnitude in increasing time windows
• Regression analysis based on the European
strong motion Data-Base (ESD, Ambraseys,
2004)
European Strong Motion Data Base*
* Ambraseys et al. (2004)
• 207 Events with 4≤MW≤7.4 (Kokaeli,1999)
• 376 three-component records
• Epicentral distance ≤ 50 km
• Low-pass filter: 3 Hz
• Magnitude bin: 0.3
Vertical
Measurement of Peak Amplitude
2-sec
H(t)=NS2(t)+EW2(t)
Horizontal
1-sec
3Hz low-pass filtered acceleration at station Bagnoli
(1980, Irpinia Eqk, Ms=6.9) – Epicentral Distance: 20 Km
Log(displacement)
Log(PGDt) vs Magnitude
P-wave
S-wave
Mean value
2-Weighted
Standard Error
Single data point
magnitude
A possible explanation
The “far-field” approximation for displacement (f=3Hz, D> 5-6 km):
moment rate
Slip-rate vs M
Active slip area vs M
Rupture kinematics
Rupture dynamics
Dynamic stress drop vs M
Rise-time vs M
The observed correlation between log(PGXt) and Magnitude would imply that
“dynamic stress release” and/or “rise-time”
scale with earthquake size in the very early stage of seismic ruptures
Summary
A high-density, wide-dynamics seismic network is being
installed in southern Italy for “regional” early-warning
applications
The system will implement a real-time eqk location method
based on an evolutionary, probabilistic approach
Early P- and S- signal amplitudes (less the 2 sec from first
arrival) correlate with magnitude (4≤Mw≤7.4) as from the
analysis of the European Strong Motion Data Base
A combination of magnitude estimations obtained by “early
P/S peak amplitudes” and “predominant periods” (Allen &
Kanamori,2003) measured at different stations as a
function of time may significantly improve the accuracy of
the earthquake size estimation in real-time procedures.
The End
Multiple Events
Each time a new pick is available,
the algorithm:
1. Temporarily associates the pick
to the current event
2. Relocates the event
3. Checks the travel-time RMS for
the maximum likelihood
hypocenter
4. If RMS < RMSthresh the pick is
definitively associated,
otherwise a new event is
declared
Vertical
Modulus of
horizontal
components
Strong Motion Data
De Natale et al., BSSA,1987
Kanamori & Rivera, BSSA, 2004
acceleration
velocity
displacement
Log(PGXt) vs Magnitude
Early P-Wave
Early S-Wave
Slide 8
Real-Time Estimation of Earthquake Location
and Magnitude for Seismic Early Warning in
Campania Region, southern Italy
A. Zollo and RISSC-Lab Research Group*
with A.Lomax (A.Lomax Scientific Software)
*
is a joint seismological research group between University of
Naples - Dept of Physics and INGV – Osservatorio Vesuviano
Work Motivation
• Development and testing of a Seismic EarlyWarning System for automated risk mitigation
actions in Campania Region
• Need for robust and reliable real-time estimates
of eqk location and magnitude to be obtained in
an evolving, continually updated form.
• Need to provide with parameter uncertainty
variation with time engineering structural
control
Historical Earthquakes
1980 Irpinia
earthquake,
Ms=6.9
Recent Seismicity
Early Warning
Network
INGV catalogue
(1981-2002)
M2.5
29 sites
Osiris 24-bit Data Logger
6 channels:
3 accelerometers
3 seismometers
(Short Period or
Broad Band)
Real time data analysis
SEW System Peculiarities
Characteristic times:
3-5 s
1.5 - 3.5 s
16 - 18 s
eqk at 4-16 km
depth
Latency
&
computation
60 km
22 - 24 s
80 km
28 - 30 s
100 km
time
To
TP first
TS target
High spatial density :
Station spacing < 15 km
Wide-Dynamics:
Unsaturated signals up to 1 g
Real-Time Earthquake Location
Basic Ideas:
Constraint from “not-yet-triggered” stations
Tracing and crossing Equal Differential
Time (EDT) surfaces
Probabilistic estimation of eqk location vs
time (Evolutionary Approach)
Real-time Evolutionary Location
• When a first station Sn triggers at tn = tnow,
we can already place limits on a pdf
volume that is likely to contain the
hypocenter (Voronoi cell). These limits
are given by conditional EDT surfaces on
which the P travel time to the first
triggering station A is equal to the traveltime to each of the not-yet-triggered
stations.
• As the current time tnow progresses we
gain the additional information that the
not-yet-triggered stations can only trigger
with tl > tnow
• When the second and later stations
trigger, we construct standard, true EDT
surfaces between each pair of the
triggered stations.These EDT surfaces
are stacked with the volume defined by
the not-yet-triggered stations to form the
current hypocentral pdf volume.
Synthetic Examples
Earthquake location probability
Seconds from first trigger
Seconds from earthquake Origin Time
Triggered stations
Earthquake location probability
Real-Time Magnitude Estimate
Basic Ideas
• Use both early P- and S-wave information based
on the high density / wide dynamics of the
network
• Correlate low-pass filtered peak amplitudes with
Magnitude in increasing time windows
• Regression analysis based on the European
strong motion Data-Base (ESD, Ambraseys,
2004)
European Strong Motion Data Base*
* Ambraseys et al. (2004)
• 207 Events with 4≤MW≤7.4 (Kokaeli,1999)
• 376 three-component records
• Epicentral distance ≤ 50 km
• Low-pass filter: 3 Hz
• Magnitude bin: 0.3
Vertical
Measurement of Peak Amplitude
2-sec
H(t)=NS2(t)+EW2(t)
Horizontal
1-sec
3Hz low-pass filtered acceleration at station Bagnoli
(1980, Irpinia Eqk, Ms=6.9) – Epicentral Distance: 20 Km
Log(displacement)
Log(PGDt) vs Magnitude
P-wave
S-wave
Mean value
2-Weighted
Standard Error
Single data point
magnitude
A possible explanation
The “far-field” approximation for displacement (f=3Hz, D> 5-6 km):
moment rate
Slip-rate vs M
Active slip area vs M
Rupture kinematics
Rupture dynamics
Dynamic stress drop vs M
Rise-time vs M
The observed correlation between log(PGXt) and Magnitude would imply that
“dynamic stress release” and/or “rise-time”
scale with earthquake size in the very early stage of seismic ruptures
Summary
A high-density, wide-dynamics seismic network is being
installed in southern Italy for “regional” early-warning
applications
The system will implement a real-time eqk location method
based on an evolutionary, probabilistic approach
Early P- and S- signal amplitudes (less the 2 sec from first
arrival) correlate with magnitude (4≤Mw≤7.4) as from the
analysis of the European Strong Motion Data Base
A combination of magnitude estimations obtained by “early
P/S peak amplitudes” and “predominant periods” (Allen &
Kanamori,2003) measured at different stations as a
function of time may significantly improve the accuracy of
the earthquake size estimation in real-time procedures.
The End
Multiple Events
Each time a new pick is available,
the algorithm:
1. Temporarily associates the pick
to the current event
2. Relocates the event
3. Checks the travel-time RMS for
the maximum likelihood
hypocenter
4. If RMS < RMSthresh the pick is
definitively associated,
otherwise a new event is
declared
Vertical
Modulus of
horizontal
components
Strong Motion Data
De Natale et al., BSSA,1987
Kanamori & Rivera, BSSA, 2004
acceleration
velocity
displacement
Log(PGXt) vs Magnitude
Early P-Wave
Early S-Wave
Slide 9
Real-Time Estimation of Earthquake Location
and Magnitude for Seismic Early Warning in
Campania Region, southern Italy
A. Zollo and RISSC-Lab Research Group*
with A.Lomax (A.Lomax Scientific Software)
*
is a joint seismological research group between University of
Naples - Dept of Physics and INGV – Osservatorio Vesuviano
Work Motivation
• Development and testing of a Seismic EarlyWarning System for automated risk mitigation
actions in Campania Region
• Need for robust and reliable real-time estimates
of eqk location and magnitude to be obtained in
an evolving, continually updated form.
• Need to provide with parameter uncertainty
variation with time engineering structural
control
Historical Earthquakes
1980 Irpinia
earthquake,
Ms=6.9
Recent Seismicity
Early Warning
Network
INGV catalogue
(1981-2002)
M2.5
29 sites
Osiris 24-bit Data Logger
6 channels:
3 accelerometers
3 seismometers
(Short Period or
Broad Band)
Real time data analysis
SEW System Peculiarities
Characteristic times:
3-5 s
1.5 - 3.5 s
16 - 18 s
eqk at 4-16 km
depth
Latency
&
computation
60 km
22 - 24 s
80 km
28 - 30 s
100 km
time
To
TP first
TS target
High spatial density :
Station spacing < 15 km
Wide-Dynamics:
Unsaturated signals up to 1 g
Real-Time Earthquake Location
Basic Ideas:
Constraint from “not-yet-triggered” stations
Tracing and crossing Equal Differential
Time (EDT) surfaces
Probabilistic estimation of eqk location vs
time (Evolutionary Approach)
Real-time Evolutionary Location
• When a first station Sn triggers at tn = tnow,
we can already place limits on a pdf
volume that is likely to contain the
hypocenter (Voronoi cell). These limits
are given by conditional EDT surfaces on
which the P travel time to the first
triggering station A is equal to the traveltime to each of the not-yet-triggered
stations.
• As the current time tnow progresses we
gain the additional information that the
not-yet-triggered stations can only trigger
with tl > tnow
• When the second and later stations
trigger, we construct standard, true EDT
surfaces between each pair of the
triggered stations.These EDT surfaces
are stacked with the volume defined by
the not-yet-triggered stations to form the
current hypocentral pdf volume.
Synthetic Examples
Earthquake location probability
Seconds from first trigger
Seconds from earthquake Origin Time
Triggered stations
Earthquake location probability
Real-Time Magnitude Estimate
Basic Ideas
• Use both early P- and S-wave information based
on the high density / wide dynamics of the
network
• Correlate low-pass filtered peak amplitudes with
Magnitude in increasing time windows
• Regression analysis based on the European
strong motion Data-Base (ESD, Ambraseys,
2004)
European Strong Motion Data Base*
* Ambraseys et al. (2004)
• 207 Events with 4≤MW≤7.4 (Kokaeli,1999)
• 376 three-component records
• Epicentral distance ≤ 50 km
• Low-pass filter: 3 Hz
• Magnitude bin: 0.3
Vertical
Measurement of Peak Amplitude
2-sec
H(t)=NS2(t)+EW2(t)
Horizontal
1-sec
3Hz low-pass filtered acceleration at station Bagnoli
(1980, Irpinia Eqk, Ms=6.9) – Epicentral Distance: 20 Km
Log(displacement)
Log(PGDt) vs Magnitude
P-wave
S-wave
Mean value
2-Weighted
Standard Error
Single data point
magnitude
A possible explanation
The “far-field” approximation for displacement (f=3Hz, D> 5-6 km):
moment rate
Slip-rate vs M
Active slip area vs M
Rupture kinematics
Rupture dynamics
Dynamic stress drop vs M
Rise-time vs M
The observed correlation between log(PGXt) and Magnitude would imply that
“dynamic stress release” and/or “rise-time”
scale with earthquake size in the very early stage of seismic ruptures
Summary
A high-density, wide-dynamics seismic network is being
installed in southern Italy for “regional” early-warning
applications
The system will implement a real-time eqk location method
based on an evolutionary, probabilistic approach
Early P- and S- signal amplitudes (less the 2 sec from first
arrival) correlate with magnitude (4≤Mw≤7.4) as from the
analysis of the European Strong Motion Data Base
A combination of magnitude estimations obtained by “early
P/S peak amplitudes” and “predominant periods” (Allen &
Kanamori,2003) measured at different stations as a
function of time may significantly improve the accuracy of
the earthquake size estimation in real-time procedures.
The End
Multiple Events
Each time a new pick is available,
the algorithm:
1. Temporarily associates the pick
to the current event
2. Relocates the event
3. Checks the travel-time RMS for
the maximum likelihood
hypocenter
4. If RMS < RMSthresh the pick is
definitively associated,
otherwise a new event is
declared
Vertical
Modulus of
horizontal
components
Strong Motion Data
De Natale et al., BSSA,1987
Kanamori & Rivera, BSSA, 2004
acceleration
velocity
displacement
Log(PGXt) vs Magnitude
Early P-Wave
Early S-Wave
Slide 10
Real-Time Estimation of Earthquake Location
and Magnitude for Seismic Early Warning in
Campania Region, southern Italy
A. Zollo and RISSC-Lab Research Group*
with A.Lomax (A.Lomax Scientific Software)
*
is a joint seismological research group between University of
Naples - Dept of Physics and INGV – Osservatorio Vesuviano
Work Motivation
• Development and testing of a Seismic EarlyWarning System for automated risk mitigation
actions in Campania Region
• Need for robust and reliable real-time estimates
of eqk location and magnitude to be obtained in
an evolving, continually updated form.
• Need to provide with parameter uncertainty
variation with time engineering structural
control
Historical Earthquakes
1980 Irpinia
earthquake,
Ms=6.9
Recent Seismicity
Early Warning
Network
INGV catalogue
(1981-2002)
M2.5
29 sites
Osiris 24-bit Data Logger
6 channels:
3 accelerometers
3 seismometers
(Short Period or
Broad Band)
Real time data analysis
SEW System Peculiarities
Characteristic times:
3-5 s
1.5 - 3.5 s
16 - 18 s
eqk at 4-16 km
depth
Latency
&
computation
60 km
22 - 24 s
80 km
28 - 30 s
100 km
time
To
TP first
TS target
High spatial density :
Station spacing < 15 km
Wide-Dynamics:
Unsaturated signals up to 1 g
Real-Time Earthquake Location
Basic Ideas:
Constraint from “not-yet-triggered” stations
Tracing and crossing Equal Differential
Time (EDT) surfaces
Probabilistic estimation of eqk location vs
time (Evolutionary Approach)
Real-time Evolutionary Location
• When a first station Sn triggers at tn = tnow,
we can already place limits on a pdf
volume that is likely to contain the
hypocenter (Voronoi cell). These limits
are given by conditional EDT surfaces on
which the P travel time to the first
triggering station A is equal to the traveltime to each of the not-yet-triggered
stations.
• As the current time tnow progresses we
gain the additional information that the
not-yet-triggered stations can only trigger
with tl > tnow
• When the second and later stations
trigger, we construct standard, true EDT
surfaces between each pair of the
triggered stations.These EDT surfaces
are stacked with the volume defined by
the not-yet-triggered stations to form the
current hypocentral pdf volume.
Synthetic Examples
Earthquake location probability
Seconds from first trigger
Seconds from earthquake Origin Time
Triggered stations
Earthquake location probability
Real-Time Magnitude Estimate
Basic Ideas
• Use both early P- and S-wave information based
on the high density / wide dynamics of the
network
• Correlate low-pass filtered peak amplitudes with
Magnitude in increasing time windows
• Regression analysis based on the European
strong motion Data-Base (ESD, Ambraseys,
2004)
European Strong Motion Data Base*
* Ambraseys et al. (2004)
• 207 Events with 4≤MW≤7.4 (Kokaeli,1999)
• 376 three-component records
• Epicentral distance ≤ 50 km
• Low-pass filter: 3 Hz
• Magnitude bin: 0.3
Vertical
Measurement of Peak Amplitude
2-sec
H(t)=NS2(t)+EW2(t)
Horizontal
1-sec
3Hz low-pass filtered acceleration at station Bagnoli
(1980, Irpinia Eqk, Ms=6.9) – Epicentral Distance: 20 Km
Log(displacement)
Log(PGDt) vs Magnitude
P-wave
S-wave
Mean value
2-Weighted
Standard Error
Single data point
magnitude
A possible explanation
The “far-field” approximation for displacement (f=3Hz, D> 5-6 km):
moment rate
Slip-rate vs M
Active slip area vs M
Rupture kinematics
Rupture dynamics
Dynamic stress drop vs M
Rise-time vs M
The observed correlation between log(PGXt) and Magnitude would imply that
“dynamic stress release” and/or “rise-time”
scale with earthquake size in the very early stage of seismic ruptures
Summary
A high-density, wide-dynamics seismic network is being
installed in southern Italy for “regional” early-warning
applications
The system will implement a real-time eqk location method
based on an evolutionary, probabilistic approach
Early P- and S- signal amplitudes (less the 2 sec from first
arrival) correlate with magnitude (4≤Mw≤7.4) as from the
analysis of the European Strong Motion Data Base
A combination of magnitude estimations obtained by “early
P/S peak amplitudes” and “predominant periods” (Allen &
Kanamori,2003) measured at different stations as a
function of time may significantly improve the accuracy of
the earthquake size estimation in real-time procedures.
The End
Multiple Events
Each time a new pick is available,
the algorithm:
1. Temporarily associates the pick
to the current event
2. Relocates the event
3. Checks the travel-time RMS for
the maximum likelihood
hypocenter
4. If RMS < RMSthresh the pick is
definitively associated,
otherwise a new event is
declared
Vertical
Modulus of
horizontal
components
Strong Motion Data
De Natale et al., BSSA,1987
Kanamori & Rivera, BSSA, 2004
acceleration
velocity
displacement
Log(PGXt) vs Magnitude
Early P-Wave
Early S-Wave
Slide 11
Real-Time Estimation of Earthquake Location
and Magnitude for Seismic Early Warning in
Campania Region, southern Italy
A. Zollo and RISSC-Lab Research Group*
with A.Lomax (A.Lomax Scientific Software)
*
is a joint seismological research group between University of
Naples - Dept of Physics and INGV – Osservatorio Vesuviano
Work Motivation
• Development and testing of a Seismic EarlyWarning System for automated risk mitigation
actions in Campania Region
• Need for robust and reliable real-time estimates
of eqk location and magnitude to be obtained in
an evolving, continually updated form.
• Need to provide with parameter uncertainty
variation with time engineering structural
control
Historical Earthquakes
1980 Irpinia
earthquake,
Ms=6.9
Recent Seismicity
Early Warning
Network
INGV catalogue
(1981-2002)
M2.5
29 sites
Osiris 24-bit Data Logger
6 channels:
3 accelerometers
3 seismometers
(Short Period or
Broad Band)
Real time data analysis
SEW System Peculiarities
Characteristic times:
3-5 s
1.5 - 3.5 s
16 - 18 s
eqk at 4-16 km
depth
Latency
&
computation
60 km
22 - 24 s
80 km
28 - 30 s
100 km
time
To
TP first
TS target
High spatial density :
Station spacing < 15 km
Wide-Dynamics:
Unsaturated signals up to 1 g
Real-Time Earthquake Location
Basic Ideas:
Constraint from “not-yet-triggered” stations
Tracing and crossing Equal Differential
Time (EDT) surfaces
Probabilistic estimation of eqk location vs
time (Evolutionary Approach)
Real-time Evolutionary Location
• When a first station Sn triggers at tn = tnow,
we can already place limits on a pdf
volume that is likely to contain the
hypocenter (Voronoi cell). These limits
are given by conditional EDT surfaces on
which the P travel time to the first
triggering station A is equal to the traveltime to each of the not-yet-triggered
stations.
• As the current time tnow progresses we
gain the additional information that the
not-yet-triggered stations can only trigger
with tl > tnow
• When the second and later stations
trigger, we construct standard, true EDT
surfaces between each pair of the
triggered stations.These EDT surfaces
are stacked with the volume defined by
the not-yet-triggered stations to form the
current hypocentral pdf volume.
Synthetic Examples
Earthquake location probability
Seconds from first trigger
Seconds from earthquake Origin Time
Triggered stations
Earthquake location probability
Real-Time Magnitude Estimate
Basic Ideas
• Use both early P- and S-wave information based
on the high density / wide dynamics of the
network
• Correlate low-pass filtered peak amplitudes with
Magnitude in increasing time windows
• Regression analysis based on the European
strong motion Data-Base (ESD, Ambraseys,
2004)
European Strong Motion Data Base*
* Ambraseys et al. (2004)
• 207 Events with 4≤MW≤7.4 (Kokaeli,1999)
• 376 three-component records
• Epicentral distance ≤ 50 km
• Low-pass filter: 3 Hz
• Magnitude bin: 0.3
Vertical
Measurement of Peak Amplitude
2-sec
H(t)=NS2(t)+EW2(t)
Horizontal
1-sec
3Hz low-pass filtered acceleration at station Bagnoli
(1980, Irpinia Eqk, Ms=6.9) – Epicentral Distance: 20 Km
Log(displacement)
Log(PGDt) vs Magnitude
P-wave
S-wave
Mean value
2-Weighted
Standard Error
Single data point
magnitude
A possible explanation
The “far-field” approximation for displacement (f=3Hz, D> 5-6 km):
moment rate
Slip-rate vs M
Active slip area vs M
Rupture kinematics
Rupture dynamics
Dynamic stress drop vs M
Rise-time vs M
The observed correlation between log(PGXt) and Magnitude would imply that
“dynamic stress release” and/or “rise-time”
scale with earthquake size in the very early stage of seismic ruptures
Summary
A high-density, wide-dynamics seismic network is being
installed in southern Italy for “regional” early-warning
applications
The system will implement a real-time eqk location method
based on an evolutionary, probabilistic approach
Early P- and S- signal amplitudes (less the 2 sec from first
arrival) correlate with magnitude (4≤Mw≤7.4) as from the
analysis of the European Strong Motion Data Base
A combination of magnitude estimations obtained by “early
P/S peak amplitudes” and “predominant periods” (Allen &
Kanamori,2003) measured at different stations as a
function of time may significantly improve the accuracy of
the earthquake size estimation in real-time procedures.
The End
Multiple Events
Each time a new pick is available,
the algorithm:
1. Temporarily associates the pick
to the current event
2. Relocates the event
3. Checks the travel-time RMS for
the maximum likelihood
hypocenter
4. If RMS < RMSthresh the pick is
definitively associated,
otherwise a new event is
declared
Vertical
Modulus of
horizontal
components
Strong Motion Data
De Natale et al., BSSA,1987
Kanamori & Rivera, BSSA, 2004
acceleration
velocity
displacement
Log(PGXt) vs Magnitude
Early P-Wave
Early S-Wave
Slide 12
Real-Time Estimation of Earthquake Location
and Magnitude for Seismic Early Warning in
Campania Region, southern Italy
A. Zollo and RISSC-Lab Research Group*
with A.Lomax (A.Lomax Scientific Software)
*
is a joint seismological research group between University of
Naples - Dept of Physics and INGV – Osservatorio Vesuviano
Work Motivation
• Development and testing of a Seismic EarlyWarning System for automated risk mitigation
actions in Campania Region
• Need for robust and reliable real-time estimates
of eqk location and magnitude to be obtained in
an evolving, continually updated form.
• Need to provide with parameter uncertainty
variation with time engineering structural
control
Historical Earthquakes
1980 Irpinia
earthquake,
Ms=6.9
Recent Seismicity
Early Warning
Network
INGV catalogue
(1981-2002)
M2.5
29 sites
Osiris 24-bit Data Logger
6 channels:
3 accelerometers
3 seismometers
(Short Period or
Broad Band)
Real time data analysis
SEW System Peculiarities
Characteristic times:
3-5 s
1.5 - 3.5 s
16 - 18 s
eqk at 4-16 km
depth
Latency
&
computation
60 km
22 - 24 s
80 km
28 - 30 s
100 km
time
To
TP first
TS target
High spatial density :
Station spacing < 15 km
Wide-Dynamics:
Unsaturated signals up to 1 g
Real-Time Earthquake Location
Basic Ideas:
Constraint from “not-yet-triggered” stations
Tracing and crossing Equal Differential
Time (EDT) surfaces
Probabilistic estimation of eqk location vs
time (Evolutionary Approach)
Real-time Evolutionary Location
• When a first station Sn triggers at tn = tnow,
we can already place limits on a pdf
volume that is likely to contain the
hypocenter (Voronoi cell). These limits
are given by conditional EDT surfaces on
which the P travel time to the first
triggering station A is equal to the traveltime to each of the not-yet-triggered
stations.
• As the current time tnow progresses we
gain the additional information that the
not-yet-triggered stations can only trigger
with tl > tnow
• When the second and later stations
trigger, we construct standard, true EDT
surfaces between each pair of the
triggered stations.These EDT surfaces
are stacked with the volume defined by
the not-yet-triggered stations to form the
current hypocentral pdf volume.
Synthetic Examples
Earthquake location probability
Seconds from first trigger
Seconds from earthquake Origin Time
Triggered stations
Earthquake location probability
Real-Time Magnitude Estimate
Basic Ideas
• Use both early P- and S-wave information based
on the high density / wide dynamics of the
network
• Correlate low-pass filtered peak amplitudes with
Magnitude in increasing time windows
• Regression analysis based on the European
strong motion Data-Base (ESD, Ambraseys,
2004)
European Strong Motion Data Base*
* Ambraseys et al. (2004)
• 207 Events with 4≤MW≤7.4 (Kokaeli,1999)
• 376 three-component records
• Epicentral distance ≤ 50 km
• Low-pass filter: 3 Hz
• Magnitude bin: 0.3
Vertical
Measurement of Peak Amplitude
2-sec
H(t)=NS2(t)+EW2(t)
Horizontal
1-sec
3Hz low-pass filtered acceleration at station Bagnoli
(1980, Irpinia Eqk, Ms=6.9) – Epicentral Distance: 20 Km
Log(displacement)
Log(PGDt) vs Magnitude
P-wave
S-wave
Mean value
2-Weighted
Standard Error
Single data point
magnitude
A possible explanation
The “far-field” approximation for displacement (f=3Hz, D> 5-6 km):
moment rate
Slip-rate vs M
Active slip area vs M
Rupture kinematics
Rupture dynamics
Dynamic stress drop vs M
Rise-time vs M
The observed correlation between log(PGXt) and Magnitude would imply that
“dynamic stress release” and/or “rise-time”
scale with earthquake size in the very early stage of seismic ruptures
Summary
A high-density, wide-dynamics seismic network is being
installed in southern Italy for “regional” early-warning
applications
The system will implement a real-time eqk location method
based on an evolutionary, probabilistic approach
Early P- and S- signal amplitudes (less the 2 sec from first
arrival) correlate with magnitude (4≤Mw≤7.4) as from the
analysis of the European Strong Motion Data Base
A combination of magnitude estimations obtained by “early
P/S peak amplitudes” and “predominant periods” (Allen &
Kanamori,2003) measured at different stations as a
function of time may significantly improve the accuracy of
the earthquake size estimation in real-time procedures.
The End
Multiple Events
Each time a new pick is available,
the algorithm:
1. Temporarily associates the pick
to the current event
2. Relocates the event
3. Checks the travel-time RMS for
the maximum likelihood
hypocenter
4. If RMS < RMSthresh the pick is
definitively associated,
otherwise a new event is
declared
Vertical
Modulus of
horizontal
components
Strong Motion Data
De Natale et al., BSSA,1987
Kanamori & Rivera, BSSA, 2004
acceleration
velocity
displacement
Log(PGXt) vs Magnitude
Early P-Wave
Early S-Wave
Slide 13
Real-Time Estimation of Earthquake Location
and Magnitude for Seismic Early Warning in
Campania Region, southern Italy
A. Zollo and RISSC-Lab Research Group*
with A.Lomax (A.Lomax Scientific Software)
*
is a joint seismological research group between University of
Naples - Dept of Physics and INGV – Osservatorio Vesuviano
Work Motivation
• Development and testing of a Seismic EarlyWarning System for automated risk mitigation
actions in Campania Region
• Need for robust and reliable real-time estimates
of eqk location and magnitude to be obtained in
an evolving, continually updated form.
• Need to provide with parameter uncertainty
variation with time engineering structural
control
Historical Earthquakes
1980 Irpinia
earthquake,
Ms=6.9
Recent Seismicity
Early Warning
Network
INGV catalogue
(1981-2002)
M2.5
29 sites
Osiris 24-bit Data Logger
6 channels:
3 accelerometers
3 seismometers
(Short Period or
Broad Band)
Real time data analysis
SEW System Peculiarities
Characteristic times:
3-5 s
1.5 - 3.5 s
16 - 18 s
eqk at 4-16 km
depth
Latency
&
computation
60 km
22 - 24 s
80 km
28 - 30 s
100 km
time
To
TP first
TS target
High spatial density :
Station spacing < 15 km
Wide-Dynamics:
Unsaturated signals up to 1 g
Real-Time Earthquake Location
Basic Ideas:
Constraint from “not-yet-triggered” stations
Tracing and crossing Equal Differential
Time (EDT) surfaces
Probabilistic estimation of eqk location vs
time (Evolutionary Approach)
Real-time Evolutionary Location
• When a first station Sn triggers at tn = tnow,
we can already place limits on a pdf
volume that is likely to contain the
hypocenter (Voronoi cell). These limits
are given by conditional EDT surfaces on
which the P travel time to the first
triggering station A is equal to the traveltime to each of the not-yet-triggered
stations.
• As the current time tnow progresses we
gain the additional information that the
not-yet-triggered stations can only trigger
with tl > tnow
• When the second and later stations
trigger, we construct standard, true EDT
surfaces between each pair of the
triggered stations.These EDT surfaces
are stacked with the volume defined by
the not-yet-triggered stations to form the
current hypocentral pdf volume.
Synthetic Examples
Earthquake location probability
Seconds from first trigger
Seconds from earthquake Origin Time
Triggered stations
Earthquake location probability
Real-Time Magnitude Estimate
Basic Ideas
• Use both early P- and S-wave information based
on the high density / wide dynamics of the
network
• Correlate low-pass filtered peak amplitudes with
Magnitude in increasing time windows
• Regression analysis based on the European
strong motion Data-Base (ESD, Ambraseys,
2004)
European Strong Motion Data Base*
* Ambraseys et al. (2004)
• 207 Events with 4≤MW≤7.4 (Kokaeli,1999)
• 376 three-component records
• Epicentral distance ≤ 50 km
• Low-pass filter: 3 Hz
• Magnitude bin: 0.3
Vertical
Measurement of Peak Amplitude
2-sec
H(t)=NS2(t)+EW2(t)
Horizontal
1-sec
3Hz low-pass filtered acceleration at station Bagnoli
(1980, Irpinia Eqk, Ms=6.9) – Epicentral Distance: 20 Km
Log(displacement)
Log(PGDt) vs Magnitude
P-wave
S-wave
Mean value
2-Weighted
Standard Error
Single data point
magnitude
A possible explanation
The “far-field” approximation for displacement (f=3Hz, D> 5-6 km):
moment rate
Slip-rate vs M
Active slip area vs M
Rupture kinematics
Rupture dynamics
Dynamic stress drop vs M
Rise-time vs M
The observed correlation between log(PGXt) and Magnitude would imply that
“dynamic stress release” and/or “rise-time”
scale with earthquake size in the very early stage of seismic ruptures
Summary
A high-density, wide-dynamics seismic network is being
installed in southern Italy for “regional” early-warning
applications
The system will implement a real-time eqk location method
based on an evolutionary, probabilistic approach
Early P- and S- signal amplitudes (less the 2 sec from first
arrival) correlate with magnitude (4≤Mw≤7.4) as from the
analysis of the European Strong Motion Data Base
A combination of magnitude estimations obtained by “early
P/S peak amplitudes” and “predominant periods” (Allen &
Kanamori,2003) measured at different stations as a
function of time may significantly improve the accuracy of
the earthquake size estimation in real-time procedures.
The End
Multiple Events
Each time a new pick is available,
the algorithm:
1. Temporarily associates the pick
to the current event
2. Relocates the event
3. Checks the travel-time RMS for
the maximum likelihood
hypocenter
4. If RMS < RMSthresh the pick is
definitively associated,
otherwise a new event is
declared
Vertical
Modulus of
horizontal
components
Strong Motion Data
De Natale et al., BSSA,1987
Kanamori & Rivera, BSSA, 2004
acceleration
velocity
displacement
Log(PGXt) vs Magnitude
Early P-Wave
Early S-Wave
Slide 14
Real-Time Estimation of Earthquake Location
and Magnitude for Seismic Early Warning in
Campania Region, southern Italy
A. Zollo and RISSC-Lab Research Group*
with A.Lomax (A.Lomax Scientific Software)
*
is a joint seismological research group between University of
Naples - Dept of Physics and INGV – Osservatorio Vesuviano
Work Motivation
• Development and testing of a Seismic EarlyWarning System for automated risk mitigation
actions in Campania Region
• Need for robust and reliable real-time estimates
of eqk location and magnitude to be obtained in
an evolving, continually updated form.
• Need to provide with parameter uncertainty
variation with time engineering structural
control
Historical Earthquakes
1980 Irpinia
earthquake,
Ms=6.9
Recent Seismicity
Early Warning
Network
INGV catalogue
(1981-2002)
M2.5
29 sites
Osiris 24-bit Data Logger
6 channels:
3 accelerometers
3 seismometers
(Short Period or
Broad Band)
Real time data analysis
SEW System Peculiarities
Characteristic times:
3-5 s
1.5 - 3.5 s
16 - 18 s
eqk at 4-16 km
depth
Latency
&
computation
60 km
22 - 24 s
80 km
28 - 30 s
100 km
time
To
TP first
TS target
High spatial density :
Station spacing < 15 km
Wide-Dynamics:
Unsaturated signals up to 1 g
Real-Time Earthquake Location
Basic Ideas:
Constraint from “not-yet-triggered” stations
Tracing and crossing Equal Differential
Time (EDT) surfaces
Probabilistic estimation of eqk location vs
time (Evolutionary Approach)
Real-time Evolutionary Location
• When a first station Sn triggers at tn = tnow,
we can already place limits on a pdf
volume that is likely to contain the
hypocenter (Voronoi cell). These limits
are given by conditional EDT surfaces on
which the P travel time to the first
triggering station A is equal to the traveltime to each of the not-yet-triggered
stations.
• As the current time tnow progresses we
gain the additional information that the
not-yet-triggered stations can only trigger
with tl > tnow
• When the second and later stations
trigger, we construct standard, true EDT
surfaces between each pair of the
triggered stations.These EDT surfaces
are stacked with the volume defined by
the not-yet-triggered stations to form the
current hypocentral pdf volume.
Synthetic Examples
Earthquake location probability
Seconds from first trigger
Seconds from earthquake Origin Time
Triggered stations
Earthquake location probability
Real-Time Magnitude Estimate
Basic Ideas
• Use both early P- and S-wave information based
on the high density / wide dynamics of the
network
• Correlate low-pass filtered peak amplitudes with
Magnitude in increasing time windows
• Regression analysis based on the European
strong motion Data-Base (ESD, Ambraseys,
2004)
European Strong Motion Data Base*
* Ambraseys et al. (2004)
• 207 Events with 4≤MW≤7.4 (Kokaeli,1999)
• 376 three-component records
• Epicentral distance ≤ 50 km
• Low-pass filter: 3 Hz
• Magnitude bin: 0.3
Vertical
Measurement of Peak Amplitude
2-sec
H(t)=NS2(t)+EW2(t)
Horizontal
1-sec
3Hz low-pass filtered acceleration at station Bagnoli
(1980, Irpinia Eqk, Ms=6.9) – Epicentral Distance: 20 Km
Log(displacement)
Log(PGDt) vs Magnitude
P-wave
S-wave
Mean value
2-Weighted
Standard Error
Single data point
magnitude
A possible explanation
The “far-field” approximation for displacement (f=3Hz, D> 5-6 km):
moment rate
Slip-rate vs M
Active slip area vs M
Rupture kinematics
Rupture dynamics
Dynamic stress drop vs M
Rise-time vs M
The observed correlation between log(PGXt) and Magnitude would imply that
“dynamic stress release” and/or “rise-time”
scale with earthquake size in the very early stage of seismic ruptures
Summary
A high-density, wide-dynamics seismic network is being
installed in southern Italy for “regional” early-warning
applications
The system will implement a real-time eqk location method
based on an evolutionary, probabilistic approach
Early P- and S- signal amplitudes (less the 2 sec from first
arrival) correlate with magnitude (4≤Mw≤7.4) as from the
analysis of the European Strong Motion Data Base
A combination of magnitude estimations obtained by “early
P/S peak amplitudes” and “predominant periods” (Allen &
Kanamori,2003) measured at different stations as a
function of time may significantly improve the accuracy of
the earthquake size estimation in real-time procedures.
The End
Multiple Events
Each time a new pick is available,
the algorithm:
1. Temporarily associates the pick
to the current event
2. Relocates the event
3. Checks the travel-time RMS for
the maximum likelihood
hypocenter
4. If RMS < RMSthresh the pick is
definitively associated,
otherwise a new event is
declared
Vertical
Modulus of
horizontal
components
Strong Motion Data
De Natale et al., BSSA,1987
Kanamori & Rivera, BSSA, 2004
acceleration
velocity
displacement
Log(PGXt) vs Magnitude
Early P-Wave
Early S-Wave
Slide 15
Real-Time Estimation of Earthquake Location
and Magnitude for Seismic Early Warning in
Campania Region, southern Italy
A. Zollo and RISSC-Lab Research Group*
with A.Lomax (A.Lomax Scientific Software)
*
is a joint seismological research group between University of
Naples - Dept of Physics and INGV – Osservatorio Vesuviano
Work Motivation
• Development and testing of a Seismic EarlyWarning System for automated risk mitigation
actions in Campania Region
• Need for robust and reliable real-time estimates
of eqk location and magnitude to be obtained in
an evolving, continually updated form.
• Need to provide with parameter uncertainty
variation with time engineering structural
control
Historical Earthquakes
1980 Irpinia
earthquake,
Ms=6.9
Recent Seismicity
Early Warning
Network
INGV catalogue
(1981-2002)
M2.5
29 sites
Osiris 24-bit Data Logger
6 channels:
3 accelerometers
3 seismometers
(Short Period or
Broad Band)
Real time data analysis
SEW System Peculiarities
Characteristic times:
3-5 s
1.5 - 3.5 s
16 - 18 s
eqk at 4-16 km
depth
Latency
&
computation
60 km
22 - 24 s
80 km
28 - 30 s
100 km
time
To
TP first
TS target
High spatial density :
Station spacing < 15 km
Wide-Dynamics:
Unsaturated signals up to 1 g
Real-Time Earthquake Location
Basic Ideas:
Constraint from “not-yet-triggered” stations
Tracing and crossing Equal Differential
Time (EDT) surfaces
Probabilistic estimation of eqk location vs
time (Evolutionary Approach)
Real-time Evolutionary Location
• When a first station Sn triggers at tn = tnow,
we can already place limits on a pdf
volume that is likely to contain the
hypocenter (Voronoi cell). These limits
are given by conditional EDT surfaces on
which the P travel time to the first
triggering station A is equal to the traveltime to each of the not-yet-triggered
stations.
• As the current time tnow progresses we
gain the additional information that the
not-yet-triggered stations can only trigger
with tl > tnow
• When the second and later stations
trigger, we construct standard, true EDT
surfaces between each pair of the
triggered stations.These EDT surfaces
are stacked with the volume defined by
the not-yet-triggered stations to form the
current hypocentral pdf volume.
Synthetic Examples
Earthquake location probability
Seconds from first trigger
Seconds from earthquake Origin Time
Triggered stations
Earthquake location probability
Real-Time Magnitude Estimate
Basic Ideas
• Use both early P- and S-wave information based
on the high density / wide dynamics of the
network
• Correlate low-pass filtered peak amplitudes with
Magnitude in increasing time windows
• Regression analysis based on the European
strong motion Data-Base (ESD, Ambraseys,
2004)
European Strong Motion Data Base*
* Ambraseys et al. (2004)
• 207 Events with 4≤MW≤7.4 (Kokaeli,1999)
• 376 three-component records
• Epicentral distance ≤ 50 km
• Low-pass filter: 3 Hz
• Magnitude bin: 0.3
Vertical
Measurement of Peak Amplitude
2-sec
H(t)=NS2(t)+EW2(t)
Horizontal
1-sec
3Hz low-pass filtered acceleration at station Bagnoli
(1980, Irpinia Eqk, Ms=6.9) – Epicentral Distance: 20 Km
Log(displacement)
Log(PGDt) vs Magnitude
P-wave
S-wave
Mean value
2-Weighted
Standard Error
Single data point
magnitude
A possible explanation
The “far-field” approximation for displacement (f=3Hz, D> 5-6 km):
moment rate
Slip-rate vs M
Active slip area vs M
Rupture kinematics
Rupture dynamics
Dynamic stress drop vs M
Rise-time vs M
The observed correlation between log(PGXt) and Magnitude would imply that
“dynamic stress release” and/or “rise-time”
scale with earthquake size in the very early stage of seismic ruptures
Summary
A high-density, wide-dynamics seismic network is being
installed in southern Italy for “regional” early-warning
applications
The system will implement a real-time eqk location method
based on an evolutionary, probabilistic approach
Early P- and S- signal amplitudes (less the 2 sec from first
arrival) correlate with magnitude (4≤Mw≤7.4) as from the
analysis of the European Strong Motion Data Base
A combination of magnitude estimations obtained by “early
P/S peak amplitudes” and “predominant periods” (Allen &
Kanamori,2003) measured at different stations as a
function of time may significantly improve the accuracy of
the earthquake size estimation in real-time procedures.
The End
Multiple Events
Each time a new pick is available,
the algorithm:
1. Temporarily associates the pick
to the current event
2. Relocates the event
3. Checks the travel-time RMS for
the maximum likelihood
hypocenter
4. If RMS < RMSthresh the pick is
definitively associated,
otherwise a new event is
declared
Vertical
Modulus of
horizontal
components
Strong Motion Data
De Natale et al., BSSA,1987
Kanamori & Rivera, BSSA, 2004
acceleration
velocity
displacement
Log(PGXt) vs Magnitude
Early P-Wave
Early S-Wave
Slide 16
Real-Time Estimation of Earthquake Location
and Magnitude for Seismic Early Warning in
Campania Region, southern Italy
A. Zollo and RISSC-Lab Research Group*
with A.Lomax (A.Lomax Scientific Software)
*
is a joint seismological research group between University of
Naples - Dept of Physics and INGV – Osservatorio Vesuviano
Work Motivation
• Development and testing of a Seismic EarlyWarning System for automated risk mitigation
actions in Campania Region
• Need for robust and reliable real-time estimates
of eqk location and magnitude to be obtained in
an evolving, continually updated form.
• Need to provide with parameter uncertainty
variation with time engineering structural
control
Historical Earthquakes
1980 Irpinia
earthquake,
Ms=6.9
Recent Seismicity
Early Warning
Network
INGV catalogue
(1981-2002)
M2.5
29 sites
Osiris 24-bit Data Logger
6 channels:
3 accelerometers
3 seismometers
(Short Period or
Broad Band)
Real time data analysis
SEW System Peculiarities
Characteristic times:
3-5 s
1.5 - 3.5 s
16 - 18 s
eqk at 4-16 km
depth
Latency
&
computation
60 km
22 - 24 s
80 km
28 - 30 s
100 km
time
To
TP first
TS target
High spatial density :
Station spacing < 15 km
Wide-Dynamics:
Unsaturated signals up to 1 g
Real-Time Earthquake Location
Basic Ideas:
Constraint from “not-yet-triggered” stations
Tracing and crossing Equal Differential
Time (EDT) surfaces
Probabilistic estimation of eqk location vs
time (Evolutionary Approach)
Real-time Evolutionary Location
• When a first station Sn triggers at tn = tnow,
we can already place limits on a pdf
volume that is likely to contain the
hypocenter (Voronoi cell). These limits
are given by conditional EDT surfaces on
which the P travel time to the first
triggering station A is equal to the traveltime to each of the not-yet-triggered
stations.
• As the current time tnow progresses we
gain the additional information that the
not-yet-triggered stations can only trigger
with tl > tnow
• When the second and later stations
trigger, we construct standard, true EDT
surfaces between each pair of the
triggered stations.These EDT surfaces
are stacked with the volume defined by
the not-yet-triggered stations to form the
current hypocentral pdf volume.
Synthetic Examples
Earthquake location probability
Seconds from first trigger
Seconds from earthquake Origin Time
Triggered stations
Earthquake location probability
Real-Time Magnitude Estimate
Basic Ideas
• Use both early P- and S-wave information based
on the high density / wide dynamics of the
network
• Correlate low-pass filtered peak amplitudes with
Magnitude in increasing time windows
• Regression analysis based on the European
strong motion Data-Base (ESD, Ambraseys,
2004)
European Strong Motion Data Base*
* Ambraseys et al. (2004)
• 207 Events with 4≤MW≤7.4 (Kokaeli,1999)
• 376 three-component records
• Epicentral distance ≤ 50 km
• Low-pass filter: 3 Hz
• Magnitude bin: 0.3
Vertical
Measurement of Peak Amplitude
2-sec
H(t)=NS2(t)+EW2(t)
Horizontal
1-sec
3Hz low-pass filtered acceleration at station Bagnoli
(1980, Irpinia Eqk, Ms=6.9) – Epicentral Distance: 20 Km
Log(displacement)
Log(PGDt) vs Magnitude
P-wave
S-wave
Mean value
2-Weighted
Standard Error
Single data point
magnitude
A possible explanation
The “far-field” approximation for displacement (f=3Hz, D> 5-6 km):
moment rate
Slip-rate vs M
Active slip area vs M
Rupture kinematics
Rupture dynamics
Dynamic stress drop vs M
Rise-time vs M
The observed correlation between log(PGXt) and Magnitude would imply that
“dynamic stress release” and/or “rise-time”
scale with earthquake size in the very early stage of seismic ruptures
Summary
A high-density, wide-dynamics seismic network is being
installed in southern Italy for “regional” early-warning
applications
The system will implement a real-time eqk location method
based on an evolutionary, probabilistic approach
Early P- and S- signal amplitudes (less the 2 sec from first
arrival) correlate with magnitude (4≤Mw≤7.4) as from the
analysis of the European Strong Motion Data Base
A combination of magnitude estimations obtained by “early
P/S peak amplitudes” and “predominant periods” (Allen &
Kanamori,2003) measured at different stations as a
function of time may significantly improve the accuracy of
the earthquake size estimation in real-time procedures.
The End
Multiple Events
Each time a new pick is available,
the algorithm:
1. Temporarily associates the pick
to the current event
2. Relocates the event
3. Checks the travel-time RMS for
the maximum likelihood
hypocenter
4. If RMS < RMSthresh the pick is
definitively associated,
otherwise a new event is
declared
Vertical
Modulus of
horizontal
components
Strong Motion Data
De Natale et al., BSSA,1987
Kanamori & Rivera, BSSA, 2004
acceleration
velocity
displacement
Log(PGXt) vs Magnitude
Early P-Wave
Early S-Wave
Slide 17
Real-Time Estimation of Earthquake Location
and Magnitude for Seismic Early Warning in
Campania Region, southern Italy
A. Zollo and RISSC-Lab Research Group*
with A.Lomax (A.Lomax Scientific Software)
*
is a joint seismological research group between University of
Naples - Dept of Physics and INGV – Osservatorio Vesuviano
Work Motivation
• Development and testing of a Seismic EarlyWarning System for automated risk mitigation
actions in Campania Region
• Need for robust and reliable real-time estimates
of eqk location and magnitude to be obtained in
an evolving, continually updated form.
• Need to provide with parameter uncertainty
variation with time engineering structural
control
Historical Earthquakes
1980 Irpinia
earthquake,
Ms=6.9
Recent Seismicity
Early Warning
Network
INGV catalogue
(1981-2002)
M2.5
29 sites
Osiris 24-bit Data Logger
6 channels:
3 accelerometers
3 seismometers
(Short Period or
Broad Band)
Real time data analysis
SEW System Peculiarities
Characteristic times:
3-5 s
1.5 - 3.5 s
16 - 18 s
eqk at 4-16 km
depth
Latency
&
computation
60 km
22 - 24 s
80 km
28 - 30 s
100 km
time
To
TP first
TS target
High spatial density :
Station spacing < 15 km
Wide-Dynamics:
Unsaturated signals up to 1 g
Real-Time Earthquake Location
Basic Ideas:
Constraint from “not-yet-triggered” stations
Tracing and crossing Equal Differential
Time (EDT) surfaces
Probabilistic estimation of eqk location vs
time (Evolutionary Approach)
Real-time Evolutionary Location
• When a first station Sn triggers at tn = tnow,
we can already place limits on a pdf
volume that is likely to contain the
hypocenter (Voronoi cell). These limits
are given by conditional EDT surfaces on
which the P travel time to the first
triggering station A is equal to the traveltime to each of the not-yet-triggered
stations.
• As the current time tnow progresses we
gain the additional information that the
not-yet-triggered stations can only trigger
with tl > tnow
• When the second and later stations
trigger, we construct standard, true EDT
surfaces between each pair of the
triggered stations.These EDT surfaces
are stacked with the volume defined by
the not-yet-triggered stations to form the
current hypocentral pdf volume.
Synthetic Examples
Earthquake location probability
Seconds from first trigger
Seconds from earthquake Origin Time
Triggered stations
Earthquake location probability
Real-Time Magnitude Estimate
Basic Ideas
• Use both early P- and S-wave information based
on the high density / wide dynamics of the
network
• Correlate low-pass filtered peak amplitudes with
Magnitude in increasing time windows
• Regression analysis based on the European
strong motion Data-Base (ESD, Ambraseys,
2004)
European Strong Motion Data Base*
* Ambraseys et al. (2004)
• 207 Events with 4≤MW≤7.4 (Kokaeli,1999)
• 376 three-component records
• Epicentral distance ≤ 50 km
• Low-pass filter: 3 Hz
• Magnitude bin: 0.3
Vertical
Measurement of Peak Amplitude
2-sec
H(t)=NS2(t)+EW2(t)
Horizontal
1-sec
3Hz low-pass filtered acceleration at station Bagnoli
(1980, Irpinia Eqk, Ms=6.9) – Epicentral Distance: 20 Km
Log(displacement)
Log(PGDt) vs Magnitude
P-wave
S-wave
Mean value
2-Weighted
Standard Error
Single data point
magnitude
A possible explanation
The “far-field” approximation for displacement (f=3Hz, D> 5-6 km):
moment rate
Slip-rate vs M
Active slip area vs M
Rupture kinematics
Rupture dynamics
Dynamic stress drop vs M
Rise-time vs M
The observed correlation between log(PGXt) and Magnitude would imply that
“dynamic stress release” and/or “rise-time”
scale with earthquake size in the very early stage of seismic ruptures
Summary
A high-density, wide-dynamics seismic network is being
installed in southern Italy for “regional” early-warning
applications
The system will implement a real-time eqk location method
based on an evolutionary, probabilistic approach
Early P- and S- signal amplitudes (less the 2 sec from first
arrival) correlate with magnitude (4≤Mw≤7.4) as from the
analysis of the European Strong Motion Data Base
A combination of magnitude estimations obtained by “early
P/S peak amplitudes” and “predominant periods” (Allen &
Kanamori,2003) measured at different stations as a
function of time may significantly improve the accuracy of
the earthquake size estimation in real-time procedures.
The End
Multiple Events
Each time a new pick is available,
the algorithm:
1. Temporarily associates the pick
to the current event
2. Relocates the event
3. Checks the travel-time RMS for
the maximum likelihood
hypocenter
4. If RMS < RMSthresh the pick is
definitively associated,
otherwise a new event is
declared
Vertical
Modulus of
horizontal
components
Strong Motion Data
De Natale et al., BSSA,1987
Kanamori & Rivera, BSSA, 2004
acceleration
velocity
displacement
Log(PGXt) vs Magnitude
Early P-Wave
Early S-Wave
Slide 18
Real-Time Estimation of Earthquake Location
and Magnitude for Seismic Early Warning in
Campania Region, southern Italy
A. Zollo and RISSC-Lab Research Group*
with A.Lomax (A.Lomax Scientific Software)
*
is a joint seismological research group between University of
Naples - Dept of Physics and INGV – Osservatorio Vesuviano
Work Motivation
• Development and testing of a Seismic EarlyWarning System for automated risk mitigation
actions in Campania Region
• Need for robust and reliable real-time estimates
of eqk location and magnitude to be obtained in
an evolving, continually updated form.
• Need to provide with parameter uncertainty
variation with time engineering structural
control
Historical Earthquakes
1980 Irpinia
earthquake,
Ms=6.9
Recent Seismicity
Early Warning
Network
INGV catalogue
(1981-2002)
M2.5
29 sites
Osiris 24-bit Data Logger
6 channels:
3 accelerometers
3 seismometers
(Short Period or
Broad Band)
Real time data analysis
SEW System Peculiarities
Characteristic times:
3-5 s
1.5 - 3.5 s
16 - 18 s
eqk at 4-16 km
depth
Latency
&
computation
60 km
22 - 24 s
80 km
28 - 30 s
100 km
time
To
TP first
TS target
High spatial density :
Station spacing < 15 km
Wide-Dynamics:
Unsaturated signals up to 1 g
Real-Time Earthquake Location
Basic Ideas:
Constraint from “not-yet-triggered” stations
Tracing and crossing Equal Differential
Time (EDT) surfaces
Probabilistic estimation of eqk location vs
time (Evolutionary Approach)
Real-time Evolutionary Location
• When a first station Sn triggers at tn = tnow,
we can already place limits on a pdf
volume that is likely to contain the
hypocenter (Voronoi cell). These limits
are given by conditional EDT surfaces on
which the P travel time to the first
triggering station A is equal to the traveltime to each of the not-yet-triggered
stations.
• As the current time tnow progresses we
gain the additional information that the
not-yet-triggered stations can only trigger
with tl > tnow
• When the second and later stations
trigger, we construct standard, true EDT
surfaces between each pair of the
triggered stations.These EDT surfaces
are stacked with the volume defined by
the not-yet-triggered stations to form the
current hypocentral pdf volume.
Synthetic Examples
Earthquake location probability
Seconds from first trigger
Seconds from earthquake Origin Time
Triggered stations
Earthquake location probability
Real-Time Magnitude Estimate
Basic Ideas
• Use both early P- and S-wave information based
on the high density / wide dynamics of the
network
• Correlate low-pass filtered peak amplitudes with
Magnitude in increasing time windows
• Regression analysis based on the European
strong motion Data-Base (ESD, Ambraseys,
2004)
European Strong Motion Data Base*
* Ambraseys et al. (2004)
• 207 Events with 4≤MW≤7.4 (Kokaeli,1999)
• 376 three-component records
• Epicentral distance ≤ 50 km
• Low-pass filter: 3 Hz
• Magnitude bin: 0.3
Vertical
Measurement of Peak Amplitude
2-sec
H(t)=NS2(t)+EW2(t)
Horizontal
1-sec
3Hz low-pass filtered acceleration at station Bagnoli
(1980, Irpinia Eqk, Ms=6.9) – Epicentral Distance: 20 Km
Log(displacement)
Log(PGDt) vs Magnitude
P-wave
S-wave
Mean value
2-Weighted
Standard Error
Single data point
magnitude
A possible explanation
The “far-field” approximation for displacement (f=3Hz, D> 5-6 km):
moment rate
Slip-rate vs M
Active slip area vs M
Rupture kinematics
Rupture dynamics
Dynamic stress drop vs M
Rise-time vs M
The observed correlation between log(PGXt) and Magnitude would imply that
“dynamic stress release” and/or “rise-time”
scale with earthquake size in the very early stage of seismic ruptures
Summary
A high-density, wide-dynamics seismic network is being
installed in southern Italy for “regional” early-warning
applications
The system will implement a real-time eqk location method
based on an evolutionary, probabilistic approach
Early P- and S- signal amplitudes (less the 2 sec from first
arrival) correlate with magnitude (4≤Mw≤7.4) as from the
analysis of the European Strong Motion Data Base
A combination of magnitude estimations obtained by “early
P/S peak amplitudes” and “predominant periods” (Allen &
Kanamori,2003) measured at different stations as a
function of time may significantly improve the accuracy of
the earthquake size estimation in real-time procedures.
The End
Multiple Events
Each time a new pick is available,
the algorithm:
1. Temporarily associates the pick
to the current event
2. Relocates the event
3. Checks the travel-time RMS for
the maximum likelihood
hypocenter
4. If RMS < RMSthresh the pick is
definitively associated,
otherwise a new event is
declared
Vertical
Modulus of
horizontal
components
Strong Motion Data
De Natale et al., BSSA,1987
Kanamori & Rivera, BSSA, 2004
acceleration
velocity
displacement
Log(PGXt) vs Magnitude
Early P-Wave
Early S-Wave
Slide 19
Real-Time Estimation of Earthquake Location
and Magnitude for Seismic Early Warning in
Campania Region, southern Italy
A. Zollo and RISSC-Lab Research Group*
with A.Lomax (A.Lomax Scientific Software)
*
is a joint seismological research group between University of
Naples - Dept of Physics and INGV – Osservatorio Vesuviano
Work Motivation
• Development and testing of a Seismic EarlyWarning System for automated risk mitigation
actions in Campania Region
• Need for robust and reliable real-time estimates
of eqk location and magnitude to be obtained in
an evolving, continually updated form.
• Need to provide with parameter uncertainty
variation with time engineering structural
control
Historical Earthquakes
1980 Irpinia
earthquake,
Ms=6.9
Recent Seismicity
Early Warning
Network
INGV catalogue
(1981-2002)
M2.5
29 sites
Osiris 24-bit Data Logger
6 channels:
3 accelerometers
3 seismometers
(Short Period or
Broad Band)
Real time data analysis
SEW System Peculiarities
Characteristic times:
3-5 s
1.5 - 3.5 s
16 - 18 s
eqk at 4-16 km
depth
Latency
&
computation
60 km
22 - 24 s
80 km
28 - 30 s
100 km
time
To
TP first
TS target
High spatial density :
Station spacing < 15 km
Wide-Dynamics:
Unsaturated signals up to 1 g
Real-Time Earthquake Location
Basic Ideas:
Constraint from “not-yet-triggered” stations
Tracing and crossing Equal Differential
Time (EDT) surfaces
Probabilistic estimation of eqk location vs
time (Evolutionary Approach)
Real-time Evolutionary Location
• When a first station Sn triggers at tn = tnow,
we can already place limits on a pdf
volume that is likely to contain the
hypocenter (Voronoi cell). These limits
are given by conditional EDT surfaces on
which the P travel time to the first
triggering station A is equal to the traveltime to each of the not-yet-triggered
stations.
• As the current time tnow progresses we
gain the additional information that the
not-yet-triggered stations can only trigger
with tl > tnow
• When the second and later stations
trigger, we construct standard, true EDT
surfaces between each pair of the
triggered stations.These EDT surfaces
are stacked with the volume defined by
the not-yet-triggered stations to form the
current hypocentral pdf volume.
Synthetic Examples
Earthquake location probability
Seconds from first trigger
Seconds from earthquake Origin Time
Triggered stations
Earthquake location probability
Real-Time Magnitude Estimate
Basic Ideas
• Use both early P- and S-wave information based
on the high density / wide dynamics of the
network
• Correlate low-pass filtered peak amplitudes with
Magnitude in increasing time windows
• Regression analysis based on the European
strong motion Data-Base (ESD, Ambraseys,
2004)
European Strong Motion Data Base*
* Ambraseys et al. (2004)
• 207 Events with 4≤MW≤7.4 (Kokaeli,1999)
• 376 three-component records
• Epicentral distance ≤ 50 km
• Low-pass filter: 3 Hz
• Magnitude bin: 0.3
Vertical
Measurement of Peak Amplitude
2-sec
H(t)=NS2(t)+EW2(t)
Horizontal
1-sec
3Hz low-pass filtered acceleration at station Bagnoli
(1980, Irpinia Eqk, Ms=6.9) – Epicentral Distance: 20 Km
Log(displacement)
Log(PGDt) vs Magnitude
P-wave
S-wave
Mean value
2-Weighted
Standard Error
Single data point
magnitude
A possible explanation
The “far-field” approximation for displacement (f=3Hz, D> 5-6 km):
moment rate
Slip-rate vs M
Active slip area vs M
Rupture kinematics
Rupture dynamics
Dynamic stress drop vs M
Rise-time vs M
The observed correlation between log(PGXt) and Magnitude would imply that
“dynamic stress release” and/or “rise-time”
scale with earthquake size in the very early stage of seismic ruptures
Summary
A high-density, wide-dynamics seismic network is being
installed in southern Italy for “regional” early-warning
applications
The system will implement a real-time eqk location method
based on an evolutionary, probabilistic approach
Early P- and S- signal amplitudes (less the 2 sec from first
arrival) correlate with magnitude (4≤Mw≤7.4) as from the
analysis of the European Strong Motion Data Base
A combination of magnitude estimations obtained by “early
P/S peak amplitudes” and “predominant periods” (Allen &
Kanamori,2003) measured at different stations as a
function of time may significantly improve the accuracy of
the earthquake size estimation in real-time procedures.
The End
Multiple Events
Each time a new pick is available,
the algorithm:
1. Temporarily associates the pick
to the current event
2. Relocates the event
3. Checks the travel-time RMS for
the maximum likelihood
hypocenter
4. If RMS < RMSthresh the pick is
definitively associated,
otherwise a new event is
declared
Vertical
Modulus of
horizontal
components
Strong Motion Data
De Natale et al., BSSA,1987
Kanamori & Rivera, BSSA, 2004
acceleration
velocity
displacement
Log(PGXt) vs Magnitude
Early P-Wave
Early S-Wave
Real-Time Estimation of Earthquake Location
and Magnitude for Seismic Early Warning in
Campania Region, southern Italy
A. Zollo and RISSC-Lab Research Group*
with A.Lomax (A.Lomax Scientific Software)
*
is a joint seismological research group between University of
Naples - Dept of Physics and INGV – Osservatorio Vesuviano
Work Motivation
• Development and testing of a Seismic EarlyWarning System for automated risk mitigation
actions in Campania Region
• Need for robust and reliable real-time estimates
of eqk location and magnitude to be obtained in
an evolving, continually updated form.
• Need to provide with parameter uncertainty
variation with time engineering structural
control
Historical Earthquakes
1980 Irpinia
earthquake,
Ms=6.9
Recent Seismicity
Early Warning
Network
INGV catalogue
(1981-2002)
M2.5
29 sites
Osiris 24-bit Data Logger
6 channels:
3 accelerometers
3 seismometers
(Short Period or
Broad Band)
Real time data analysis
SEW System Peculiarities
Characteristic times:
3-5 s
1.5 - 3.5 s
16 - 18 s
eqk at 4-16 km
depth
Latency
&
computation
60 km
22 - 24 s
80 km
28 - 30 s
100 km
time
To
TP first
TS target
High spatial density :
Station spacing < 15 km
Wide-Dynamics:
Unsaturated signals up to 1 g
Real-Time Earthquake Location
Basic Ideas:
Constraint from “not-yet-triggered” stations
Tracing and crossing Equal Differential
Time (EDT) surfaces
Probabilistic estimation of eqk location vs
time (Evolutionary Approach)
Real-time Evolutionary Location
• When a first station Sn triggers at tn = tnow,
we can already place limits on a pdf
volume that is likely to contain the
hypocenter (Voronoi cell). These limits
are given by conditional EDT surfaces on
which the P travel time to the first
triggering station A is equal to the traveltime to each of the not-yet-triggered
stations.
• As the current time tnow progresses we
gain the additional information that the
not-yet-triggered stations can only trigger
with tl > tnow
• When the second and later stations
trigger, we construct standard, true EDT
surfaces between each pair of the
triggered stations.These EDT surfaces
are stacked with the volume defined by
the not-yet-triggered stations to form the
current hypocentral pdf volume.
Synthetic Examples
Earthquake location probability
Seconds from first trigger
Seconds from earthquake Origin Time
Triggered stations
Earthquake location probability
Real-Time Magnitude Estimate
Basic Ideas
• Use both early P- and S-wave information based
on the high density / wide dynamics of the
network
• Correlate low-pass filtered peak amplitudes with
Magnitude in increasing time windows
• Regression analysis based on the European
strong motion Data-Base (ESD, Ambraseys,
2004)
European Strong Motion Data Base*
* Ambraseys et al. (2004)
• 207 Events with 4≤MW≤7.4 (Kokaeli,1999)
• 376 three-component records
• Epicentral distance ≤ 50 km
• Low-pass filter: 3 Hz
• Magnitude bin: 0.3
Vertical
Measurement of Peak Amplitude
2-sec
H(t)=NS2(t)+EW2(t)
Horizontal
1-sec
3Hz low-pass filtered acceleration at station Bagnoli
(1980, Irpinia Eqk, Ms=6.9) – Epicentral Distance: 20 Km
Log(displacement)
Log(PGDt) vs Magnitude
P-wave
S-wave
Mean value
2-Weighted
Standard Error
Single data point
magnitude
A possible explanation
The “far-field” approximation for displacement (f=3Hz, D> 5-6 km):
moment rate
Slip-rate vs M
Active slip area vs M
Rupture kinematics
Rupture dynamics
Dynamic stress drop vs M
Rise-time vs M
The observed correlation between log(PGXt) and Magnitude would imply that
“dynamic stress release” and/or “rise-time”
scale with earthquake size in the very early stage of seismic ruptures
Summary
A high-density, wide-dynamics seismic network is being
installed in southern Italy for “regional” early-warning
applications
The system will implement a real-time eqk location method
based on an evolutionary, probabilistic approach
Early P- and S- signal amplitudes (less the 2 sec from first
arrival) correlate with magnitude (4≤Mw≤7.4) as from the
analysis of the European Strong Motion Data Base
A combination of magnitude estimations obtained by “early
P/S peak amplitudes” and “predominant periods” (Allen &
Kanamori,2003) measured at different stations as a
function of time may significantly improve the accuracy of
the earthquake size estimation in real-time procedures.
The End
Multiple Events
Each time a new pick is available,
the algorithm:
1. Temporarily associates the pick
to the current event
2. Relocates the event
3. Checks the travel-time RMS for
the maximum likelihood
hypocenter
4. If RMS < RMSthresh the pick is
definitively associated,
otherwise a new event is
declared
Vertical
Modulus of
horizontal
components
Strong Motion Data
De Natale et al., BSSA,1987
Kanamori & Rivera, BSSA, 2004
acceleration
velocity
displacement
Log(PGXt) vs Magnitude
Early P-Wave
Early S-Wave
Slide 2
Real-Time Estimation of Earthquake Location
and Magnitude for Seismic Early Warning in
Campania Region, southern Italy
A. Zollo and RISSC-Lab Research Group*
with A.Lomax (A.Lomax Scientific Software)
*
is a joint seismological research group between University of
Naples - Dept of Physics and INGV – Osservatorio Vesuviano
Work Motivation
• Development and testing of a Seismic EarlyWarning System for automated risk mitigation
actions in Campania Region
• Need for robust and reliable real-time estimates
of eqk location and magnitude to be obtained in
an evolving, continually updated form.
• Need to provide with parameter uncertainty
variation with time engineering structural
control
Historical Earthquakes
1980 Irpinia
earthquake,
Ms=6.9
Recent Seismicity
Early Warning
Network
INGV catalogue
(1981-2002)
M2.5
29 sites
Osiris 24-bit Data Logger
6 channels:
3 accelerometers
3 seismometers
(Short Period or
Broad Band)
Real time data analysis
SEW System Peculiarities
Characteristic times:
3-5 s
1.5 - 3.5 s
16 - 18 s
eqk at 4-16 km
depth
Latency
&
computation
60 km
22 - 24 s
80 km
28 - 30 s
100 km
time
To
TP first
TS target
High spatial density :
Station spacing < 15 km
Wide-Dynamics:
Unsaturated signals up to 1 g
Real-Time Earthquake Location
Basic Ideas:
Constraint from “not-yet-triggered” stations
Tracing and crossing Equal Differential
Time (EDT) surfaces
Probabilistic estimation of eqk location vs
time (Evolutionary Approach)
Real-time Evolutionary Location
• When a first station Sn triggers at tn = tnow,
we can already place limits on a pdf
volume that is likely to contain the
hypocenter (Voronoi cell). These limits
are given by conditional EDT surfaces on
which the P travel time to the first
triggering station A is equal to the traveltime to each of the not-yet-triggered
stations.
• As the current time tnow progresses we
gain the additional information that the
not-yet-triggered stations can only trigger
with tl > tnow
• When the second and later stations
trigger, we construct standard, true EDT
surfaces between each pair of the
triggered stations.These EDT surfaces
are stacked with the volume defined by
the not-yet-triggered stations to form the
current hypocentral pdf volume.
Synthetic Examples
Earthquake location probability
Seconds from first trigger
Seconds from earthquake Origin Time
Triggered stations
Earthquake location probability
Real-Time Magnitude Estimate
Basic Ideas
• Use both early P- and S-wave information based
on the high density / wide dynamics of the
network
• Correlate low-pass filtered peak amplitudes with
Magnitude in increasing time windows
• Regression analysis based on the European
strong motion Data-Base (ESD, Ambraseys,
2004)
European Strong Motion Data Base*
* Ambraseys et al. (2004)
• 207 Events with 4≤MW≤7.4 (Kokaeli,1999)
• 376 three-component records
• Epicentral distance ≤ 50 km
• Low-pass filter: 3 Hz
• Magnitude bin: 0.3
Vertical
Measurement of Peak Amplitude
2-sec
H(t)=NS2(t)+EW2(t)
Horizontal
1-sec
3Hz low-pass filtered acceleration at station Bagnoli
(1980, Irpinia Eqk, Ms=6.9) – Epicentral Distance: 20 Km
Log(displacement)
Log(PGDt) vs Magnitude
P-wave
S-wave
Mean value
2-Weighted
Standard Error
Single data point
magnitude
A possible explanation
The “far-field” approximation for displacement (f=3Hz, D> 5-6 km):
moment rate
Slip-rate vs M
Active slip area vs M
Rupture kinematics
Rupture dynamics
Dynamic stress drop vs M
Rise-time vs M
The observed correlation between log(PGXt) and Magnitude would imply that
“dynamic stress release” and/or “rise-time”
scale with earthquake size in the very early stage of seismic ruptures
Summary
A high-density, wide-dynamics seismic network is being
installed in southern Italy for “regional” early-warning
applications
The system will implement a real-time eqk location method
based on an evolutionary, probabilistic approach
Early P- and S- signal amplitudes (less the 2 sec from first
arrival) correlate with magnitude (4≤Mw≤7.4) as from the
analysis of the European Strong Motion Data Base
A combination of magnitude estimations obtained by “early
P/S peak amplitudes” and “predominant periods” (Allen &
Kanamori,2003) measured at different stations as a
function of time may significantly improve the accuracy of
the earthquake size estimation in real-time procedures.
The End
Multiple Events
Each time a new pick is available,
the algorithm:
1. Temporarily associates the pick
to the current event
2. Relocates the event
3. Checks the travel-time RMS for
the maximum likelihood
hypocenter
4. If RMS < RMSthresh the pick is
definitively associated,
otherwise a new event is
declared
Vertical
Modulus of
horizontal
components
Strong Motion Data
De Natale et al., BSSA,1987
Kanamori & Rivera, BSSA, 2004
acceleration
velocity
displacement
Log(PGXt) vs Magnitude
Early P-Wave
Early S-Wave
Slide 3
Real-Time Estimation of Earthquake Location
and Magnitude for Seismic Early Warning in
Campania Region, southern Italy
A. Zollo and RISSC-Lab Research Group*
with A.Lomax (A.Lomax Scientific Software)
*
is a joint seismological research group between University of
Naples - Dept of Physics and INGV – Osservatorio Vesuviano
Work Motivation
• Development and testing of a Seismic EarlyWarning System for automated risk mitigation
actions in Campania Region
• Need for robust and reliable real-time estimates
of eqk location and magnitude to be obtained in
an evolving, continually updated form.
• Need to provide with parameter uncertainty
variation with time engineering structural
control
Historical Earthquakes
1980 Irpinia
earthquake,
Ms=6.9
Recent Seismicity
Early Warning
Network
INGV catalogue
(1981-2002)
M2.5
29 sites
Osiris 24-bit Data Logger
6 channels:
3 accelerometers
3 seismometers
(Short Period or
Broad Band)
Real time data analysis
SEW System Peculiarities
Characteristic times:
3-5 s
1.5 - 3.5 s
16 - 18 s
eqk at 4-16 km
depth
Latency
&
computation
60 km
22 - 24 s
80 km
28 - 30 s
100 km
time
To
TP first
TS target
High spatial density :
Station spacing < 15 km
Wide-Dynamics:
Unsaturated signals up to 1 g
Real-Time Earthquake Location
Basic Ideas:
Constraint from “not-yet-triggered” stations
Tracing and crossing Equal Differential
Time (EDT) surfaces
Probabilistic estimation of eqk location vs
time (Evolutionary Approach)
Real-time Evolutionary Location
• When a first station Sn triggers at tn = tnow,
we can already place limits on a pdf
volume that is likely to contain the
hypocenter (Voronoi cell). These limits
are given by conditional EDT surfaces on
which the P travel time to the first
triggering station A is equal to the traveltime to each of the not-yet-triggered
stations.
• As the current time tnow progresses we
gain the additional information that the
not-yet-triggered stations can only trigger
with tl > tnow
• When the second and later stations
trigger, we construct standard, true EDT
surfaces between each pair of the
triggered stations.These EDT surfaces
are stacked with the volume defined by
the not-yet-triggered stations to form the
current hypocentral pdf volume.
Synthetic Examples
Earthquake location probability
Seconds from first trigger
Seconds from earthquake Origin Time
Triggered stations
Earthquake location probability
Real-Time Magnitude Estimate
Basic Ideas
• Use both early P- and S-wave information based
on the high density / wide dynamics of the
network
• Correlate low-pass filtered peak amplitudes with
Magnitude in increasing time windows
• Regression analysis based on the European
strong motion Data-Base (ESD, Ambraseys,
2004)
European Strong Motion Data Base*
* Ambraseys et al. (2004)
• 207 Events with 4≤MW≤7.4 (Kokaeli,1999)
• 376 three-component records
• Epicentral distance ≤ 50 km
• Low-pass filter: 3 Hz
• Magnitude bin: 0.3
Vertical
Measurement of Peak Amplitude
2-sec
H(t)=NS2(t)+EW2(t)
Horizontal
1-sec
3Hz low-pass filtered acceleration at station Bagnoli
(1980, Irpinia Eqk, Ms=6.9) – Epicentral Distance: 20 Km
Log(displacement)
Log(PGDt) vs Magnitude
P-wave
S-wave
Mean value
2-Weighted
Standard Error
Single data point
magnitude
A possible explanation
The “far-field” approximation for displacement (f=3Hz, D> 5-6 km):
moment rate
Slip-rate vs M
Active slip area vs M
Rupture kinematics
Rupture dynamics
Dynamic stress drop vs M
Rise-time vs M
The observed correlation between log(PGXt) and Magnitude would imply that
“dynamic stress release” and/or “rise-time”
scale with earthquake size in the very early stage of seismic ruptures
Summary
A high-density, wide-dynamics seismic network is being
installed in southern Italy for “regional” early-warning
applications
The system will implement a real-time eqk location method
based on an evolutionary, probabilistic approach
Early P- and S- signal amplitudes (less the 2 sec from first
arrival) correlate with magnitude (4≤Mw≤7.4) as from the
analysis of the European Strong Motion Data Base
A combination of magnitude estimations obtained by “early
P/S peak amplitudes” and “predominant periods” (Allen &
Kanamori,2003) measured at different stations as a
function of time may significantly improve the accuracy of
the earthquake size estimation in real-time procedures.
The End
Multiple Events
Each time a new pick is available,
the algorithm:
1. Temporarily associates the pick
to the current event
2. Relocates the event
3. Checks the travel-time RMS for
the maximum likelihood
hypocenter
4. If RMS < RMSthresh the pick is
definitively associated,
otherwise a new event is
declared
Vertical
Modulus of
horizontal
components
Strong Motion Data
De Natale et al., BSSA,1987
Kanamori & Rivera, BSSA, 2004
acceleration
velocity
displacement
Log(PGXt) vs Magnitude
Early P-Wave
Early S-Wave
Slide 4
Real-Time Estimation of Earthquake Location
and Magnitude for Seismic Early Warning in
Campania Region, southern Italy
A. Zollo and RISSC-Lab Research Group*
with A.Lomax (A.Lomax Scientific Software)
*
is a joint seismological research group between University of
Naples - Dept of Physics and INGV – Osservatorio Vesuviano
Work Motivation
• Development and testing of a Seismic EarlyWarning System for automated risk mitigation
actions in Campania Region
• Need for robust and reliable real-time estimates
of eqk location and magnitude to be obtained in
an evolving, continually updated form.
• Need to provide with parameter uncertainty
variation with time engineering structural
control
Historical Earthquakes
1980 Irpinia
earthquake,
Ms=6.9
Recent Seismicity
Early Warning
Network
INGV catalogue
(1981-2002)
M2.5
29 sites
Osiris 24-bit Data Logger
6 channels:
3 accelerometers
3 seismometers
(Short Period or
Broad Band)
Real time data analysis
SEW System Peculiarities
Characteristic times:
3-5 s
1.5 - 3.5 s
16 - 18 s
eqk at 4-16 km
depth
Latency
&
computation
60 km
22 - 24 s
80 km
28 - 30 s
100 km
time
To
TP first
TS target
High spatial density :
Station spacing < 15 km
Wide-Dynamics:
Unsaturated signals up to 1 g
Real-Time Earthquake Location
Basic Ideas:
Constraint from “not-yet-triggered” stations
Tracing and crossing Equal Differential
Time (EDT) surfaces
Probabilistic estimation of eqk location vs
time (Evolutionary Approach)
Real-time Evolutionary Location
• When a first station Sn triggers at tn = tnow,
we can already place limits on a pdf
volume that is likely to contain the
hypocenter (Voronoi cell). These limits
are given by conditional EDT surfaces on
which the P travel time to the first
triggering station A is equal to the traveltime to each of the not-yet-triggered
stations.
• As the current time tnow progresses we
gain the additional information that the
not-yet-triggered stations can only trigger
with tl > tnow
• When the second and later stations
trigger, we construct standard, true EDT
surfaces between each pair of the
triggered stations.These EDT surfaces
are stacked with the volume defined by
the not-yet-triggered stations to form the
current hypocentral pdf volume.
Synthetic Examples
Earthquake location probability
Seconds from first trigger
Seconds from earthquake Origin Time
Triggered stations
Earthquake location probability
Real-Time Magnitude Estimate
Basic Ideas
• Use both early P- and S-wave information based
on the high density / wide dynamics of the
network
• Correlate low-pass filtered peak amplitudes with
Magnitude in increasing time windows
• Regression analysis based on the European
strong motion Data-Base (ESD, Ambraseys,
2004)
European Strong Motion Data Base*
* Ambraseys et al. (2004)
• 207 Events with 4≤MW≤7.4 (Kokaeli,1999)
• 376 three-component records
• Epicentral distance ≤ 50 km
• Low-pass filter: 3 Hz
• Magnitude bin: 0.3
Vertical
Measurement of Peak Amplitude
2-sec
H(t)=NS2(t)+EW2(t)
Horizontal
1-sec
3Hz low-pass filtered acceleration at station Bagnoli
(1980, Irpinia Eqk, Ms=6.9) – Epicentral Distance: 20 Km
Log(displacement)
Log(PGDt) vs Magnitude
P-wave
S-wave
Mean value
2-Weighted
Standard Error
Single data point
magnitude
A possible explanation
The “far-field” approximation for displacement (f=3Hz, D> 5-6 km):
moment rate
Slip-rate vs M
Active slip area vs M
Rupture kinematics
Rupture dynamics
Dynamic stress drop vs M
Rise-time vs M
The observed correlation between log(PGXt) and Magnitude would imply that
“dynamic stress release” and/or “rise-time”
scale with earthquake size in the very early stage of seismic ruptures
Summary
A high-density, wide-dynamics seismic network is being
installed in southern Italy for “regional” early-warning
applications
The system will implement a real-time eqk location method
based on an evolutionary, probabilistic approach
Early P- and S- signal amplitudes (less the 2 sec from first
arrival) correlate with magnitude (4≤Mw≤7.4) as from the
analysis of the European Strong Motion Data Base
A combination of magnitude estimations obtained by “early
P/S peak amplitudes” and “predominant periods” (Allen &
Kanamori,2003) measured at different stations as a
function of time may significantly improve the accuracy of
the earthquake size estimation in real-time procedures.
The End
Multiple Events
Each time a new pick is available,
the algorithm:
1. Temporarily associates the pick
to the current event
2. Relocates the event
3. Checks the travel-time RMS for
the maximum likelihood
hypocenter
4. If RMS < RMSthresh the pick is
definitively associated,
otherwise a new event is
declared
Vertical
Modulus of
horizontal
components
Strong Motion Data
De Natale et al., BSSA,1987
Kanamori & Rivera, BSSA, 2004
acceleration
velocity
displacement
Log(PGXt) vs Magnitude
Early P-Wave
Early S-Wave
Slide 5
Real-Time Estimation of Earthquake Location
and Magnitude for Seismic Early Warning in
Campania Region, southern Italy
A. Zollo and RISSC-Lab Research Group*
with A.Lomax (A.Lomax Scientific Software)
*
is a joint seismological research group between University of
Naples - Dept of Physics and INGV – Osservatorio Vesuviano
Work Motivation
• Development and testing of a Seismic EarlyWarning System for automated risk mitigation
actions in Campania Region
• Need for robust and reliable real-time estimates
of eqk location and magnitude to be obtained in
an evolving, continually updated form.
• Need to provide with parameter uncertainty
variation with time engineering structural
control
Historical Earthquakes
1980 Irpinia
earthquake,
Ms=6.9
Recent Seismicity
Early Warning
Network
INGV catalogue
(1981-2002)
M2.5
29 sites
Osiris 24-bit Data Logger
6 channels:
3 accelerometers
3 seismometers
(Short Period or
Broad Band)
Real time data analysis
SEW System Peculiarities
Characteristic times:
3-5 s
1.5 - 3.5 s
16 - 18 s
eqk at 4-16 km
depth
Latency
&
computation
60 km
22 - 24 s
80 km
28 - 30 s
100 km
time
To
TP first
TS target
High spatial density :
Station spacing < 15 km
Wide-Dynamics:
Unsaturated signals up to 1 g
Real-Time Earthquake Location
Basic Ideas:
Constraint from “not-yet-triggered” stations
Tracing and crossing Equal Differential
Time (EDT) surfaces
Probabilistic estimation of eqk location vs
time (Evolutionary Approach)
Real-time Evolutionary Location
• When a first station Sn triggers at tn = tnow,
we can already place limits on a pdf
volume that is likely to contain the
hypocenter (Voronoi cell). These limits
are given by conditional EDT surfaces on
which the P travel time to the first
triggering station A is equal to the traveltime to each of the not-yet-triggered
stations.
• As the current time tnow progresses we
gain the additional information that the
not-yet-triggered stations can only trigger
with tl > tnow
• When the second and later stations
trigger, we construct standard, true EDT
surfaces between each pair of the
triggered stations.These EDT surfaces
are stacked with the volume defined by
the not-yet-triggered stations to form the
current hypocentral pdf volume.
Synthetic Examples
Earthquake location probability
Seconds from first trigger
Seconds from earthquake Origin Time
Triggered stations
Earthquake location probability
Real-Time Magnitude Estimate
Basic Ideas
• Use both early P- and S-wave information based
on the high density / wide dynamics of the
network
• Correlate low-pass filtered peak amplitudes with
Magnitude in increasing time windows
• Regression analysis based on the European
strong motion Data-Base (ESD, Ambraseys,
2004)
European Strong Motion Data Base*
* Ambraseys et al. (2004)
• 207 Events with 4≤MW≤7.4 (Kokaeli,1999)
• 376 three-component records
• Epicentral distance ≤ 50 km
• Low-pass filter: 3 Hz
• Magnitude bin: 0.3
Vertical
Measurement of Peak Amplitude
2-sec
H(t)=NS2(t)+EW2(t)
Horizontal
1-sec
3Hz low-pass filtered acceleration at station Bagnoli
(1980, Irpinia Eqk, Ms=6.9) – Epicentral Distance: 20 Km
Log(displacement)
Log(PGDt) vs Magnitude
P-wave
S-wave
Mean value
2-Weighted
Standard Error
Single data point
magnitude
A possible explanation
The “far-field” approximation for displacement (f=3Hz, D> 5-6 km):
moment rate
Slip-rate vs M
Active slip area vs M
Rupture kinematics
Rupture dynamics
Dynamic stress drop vs M
Rise-time vs M
The observed correlation between log(PGXt) and Magnitude would imply that
“dynamic stress release” and/or “rise-time”
scale with earthquake size in the very early stage of seismic ruptures
Summary
A high-density, wide-dynamics seismic network is being
installed in southern Italy for “regional” early-warning
applications
The system will implement a real-time eqk location method
based on an evolutionary, probabilistic approach
Early P- and S- signal amplitudes (less the 2 sec from first
arrival) correlate with magnitude (4≤Mw≤7.4) as from the
analysis of the European Strong Motion Data Base
A combination of magnitude estimations obtained by “early
P/S peak amplitudes” and “predominant periods” (Allen &
Kanamori,2003) measured at different stations as a
function of time may significantly improve the accuracy of
the earthquake size estimation in real-time procedures.
The End
Multiple Events
Each time a new pick is available,
the algorithm:
1. Temporarily associates the pick
to the current event
2. Relocates the event
3. Checks the travel-time RMS for
the maximum likelihood
hypocenter
4. If RMS < RMSthresh the pick is
definitively associated,
otherwise a new event is
declared
Vertical
Modulus of
horizontal
components
Strong Motion Data
De Natale et al., BSSA,1987
Kanamori & Rivera, BSSA, 2004
acceleration
velocity
displacement
Log(PGXt) vs Magnitude
Early P-Wave
Early S-Wave
Slide 6
Real-Time Estimation of Earthquake Location
and Magnitude for Seismic Early Warning in
Campania Region, southern Italy
A. Zollo and RISSC-Lab Research Group*
with A.Lomax (A.Lomax Scientific Software)
*
is a joint seismological research group between University of
Naples - Dept of Physics and INGV – Osservatorio Vesuviano
Work Motivation
• Development and testing of a Seismic EarlyWarning System for automated risk mitigation
actions in Campania Region
• Need for robust and reliable real-time estimates
of eqk location and magnitude to be obtained in
an evolving, continually updated form.
• Need to provide with parameter uncertainty
variation with time engineering structural
control
Historical Earthquakes
1980 Irpinia
earthquake,
Ms=6.9
Recent Seismicity
Early Warning
Network
INGV catalogue
(1981-2002)
M2.5
29 sites
Osiris 24-bit Data Logger
6 channels:
3 accelerometers
3 seismometers
(Short Period or
Broad Band)
Real time data analysis
SEW System Peculiarities
Characteristic times:
3-5 s
1.5 - 3.5 s
16 - 18 s
eqk at 4-16 km
depth
Latency
&
computation
60 km
22 - 24 s
80 km
28 - 30 s
100 km
time
To
TP first
TS target
High spatial density :
Station spacing < 15 km
Wide-Dynamics:
Unsaturated signals up to 1 g
Real-Time Earthquake Location
Basic Ideas:
Constraint from “not-yet-triggered” stations
Tracing and crossing Equal Differential
Time (EDT) surfaces
Probabilistic estimation of eqk location vs
time (Evolutionary Approach)
Real-time Evolutionary Location
• When a first station Sn triggers at tn = tnow,
we can already place limits on a pdf
volume that is likely to contain the
hypocenter (Voronoi cell). These limits
are given by conditional EDT surfaces on
which the P travel time to the first
triggering station A is equal to the traveltime to each of the not-yet-triggered
stations.
• As the current time tnow progresses we
gain the additional information that the
not-yet-triggered stations can only trigger
with tl > tnow
• When the second and later stations
trigger, we construct standard, true EDT
surfaces between each pair of the
triggered stations.These EDT surfaces
are stacked with the volume defined by
the not-yet-triggered stations to form the
current hypocentral pdf volume.
Synthetic Examples
Earthquake location probability
Seconds from first trigger
Seconds from earthquake Origin Time
Triggered stations
Earthquake location probability
Real-Time Magnitude Estimate
Basic Ideas
• Use both early P- and S-wave information based
on the high density / wide dynamics of the
network
• Correlate low-pass filtered peak amplitudes with
Magnitude in increasing time windows
• Regression analysis based on the European
strong motion Data-Base (ESD, Ambraseys,
2004)
European Strong Motion Data Base*
* Ambraseys et al. (2004)
• 207 Events with 4≤MW≤7.4 (Kokaeli,1999)
• 376 three-component records
• Epicentral distance ≤ 50 km
• Low-pass filter: 3 Hz
• Magnitude bin: 0.3
Vertical
Measurement of Peak Amplitude
2-sec
H(t)=NS2(t)+EW2(t)
Horizontal
1-sec
3Hz low-pass filtered acceleration at station Bagnoli
(1980, Irpinia Eqk, Ms=6.9) – Epicentral Distance: 20 Km
Log(displacement)
Log(PGDt) vs Magnitude
P-wave
S-wave
Mean value
2-Weighted
Standard Error
Single data point
magnitude
A possible explanation
The “far-field” approximation for displacement (f=3Hz, D> 5-6 km):
moment rate
Slip-rate vs M
Active slip area vs M
Rupture kinematics
Rupture dynamics
Dynamic stress drop vs M
Rise-time vs M
The observed correlation between log(PGXt) and Magnitude would imply that
“dynamic stress release” and/or “rise-time”
scale with earthquake size in the very early stage of seismic ruptures
Summary
A high-density, wide-dynamics seismic network is being
installed in southern Italy for “regional” early-warning
applications
The system will implement a real-time eqk location method
based on an evolutionary, probabilistic approach
Early P- and S- signal amplitudes (less the 2 sec from first
arrival) correlate with magnitude (4≤Mw≤7.4) as from the
analysis of the European Strong Motion Data Base
A combination of magnitude estimations obtained by “early
P/S peak amplitudes” and “predominant periods” (Allen &
Kanamori,2003) measured at different stations as a
function of time may significantly improve the accuracy of
the earthquake size estimation in real-time procedures.
The End
Multiple Events
Each time a new pick is available,
the algorithm:
1. Temporarily associates the pick
to the current event
2. Relocates the event
3. Checks the travel-time RMS for
the maximum likelihood
hypocenter
4. If RMS < RMSthresh the pick is
definitively associated,
otherwise a new event is
declared
Vertical
Modulus of
horizontal
components
Strong Motion Data
De Natale et al., BSSA,1987
Kanamori & Rivera, BSSA, 2004
acceleration
velocity
displacement
Log(PGXt) vs Magnitude
Early P-Wave
Early S-Wave
Slide 7
Real-Time Estimation of Earthquake Location
and Magnitude for Seismic Early Warning in
Campania Region, southern Italy
A. Zollo and RISSC-Lab Research Group*
with A.Lomax (A.Lomax Scientific Software)
*
is a joint seismological research group between University of
Naples - Dept of Physics and INGV – Osservatorio Vesuviano
Work Motivation
• Development and testing of a Seismic EarlyWarning System for automated risk mitigation
actions in Campania Region
• Need for robust and reliable real-time estimates
of eqk location and magnitude to be obtained in
an evolving, continually updated form.
• Need to provide with parameter uncertainty
variation with time engineering structural
control
Historical Earthquakes
1980 Irpinia
earthquake,
Ms=6.9
Recent Seismicity
Early Warning
Network
INGV catalogue
(1981-2002)
M2.5
29 sites
Osiris 24-bit Data Logger
6 channels:
3 accelerometers
3 seismometers
(Short Period or
Broad Band)
Real time data analysis
SEW System Peculiarities
Characteristic times:
3-5 s
1.5 - 3.5 s
16 - 18 s
eqk at 4-16 km
depth
Latency
&
computation
60 km
22 - 24 s
80 km
28 - 30 s
100 km
time
To
TP first
TS target
High spatial density :
Station spacing < 15 km
Wide-Dynamics:
Unsaturated signals up to 1 g
Real-Time Earthquake Location
Basic Ideas:
Constraint from “not-yet-triggered” stations
Tracing and crossing Equal Differential
Time (EDT) surfaces
Probabilistic estimation of eqk location vs
time (Evolutionary Approach)
Real-time Evolutionary Location
• When a first station Sn triggers at tn = tnow,
we can already place limits on a pdf
volume that is likely to contain the
hypocenter (Voronoi cell). These limits
are given by conditional EDT surfaces on
which the P travel time to the first
triggering station A is equal to the traveltime to each of the not-yet-triggered
stations.
• As the current time tnow progresses we
gain the additional information that the
not-yet-triggered stations can only trigger
with tl > tnow
• When the second and later stations
trigger, we construct standard, true EDT
surfaces between each pair of the
triggered stations.These EDT surfaces
are stacked with the volume defined by
the not-yet-triggered stations to form the
current hypocentral pdf volume.
Synthetic Examples
Earthquake location probability
Seconds from first trigger
Seconds from earthquake Origin Time
Triggered stations
Earthquake location probability
Real-Time Magnitude Estimate
Basic Ideas
• Use both early P- and S-wave information based
on the high density / wide dynamics of the
network
• Correlate low-pass filtered peak amplitudes with
Magnitude in increasing time windows
• Regression analysis based on the European
strong motion Data-Base (ESD, Ambraseys,
2004)
European Strong Motion Data Base*
* Ambraseys et al. (2004)
• 207 Events with 4≤MW≤7.4 (Kokaeli,1999)
• 376 three-component records
• Epicentral distance ≤ 50 km
• Low-pass filter: 3 Hz
• Magnitude bin: 0.3
Vertical
Measurement of Peak Amplitude
2-sec
H(t)=NS2(t)+EW2(t)
Horizontal
1-sec
3Hz low-pass filtered acceleration at station Bagnoli
(1980, Irpinia Eqk, Ms=6.9) – Epicentral Distance: 20 Km
Log(displacement)
Log(PGDt) vs Magnitude
P-wave
S-wave
Mean value
2-Weighted
Standard Error
Single data point
magnitude
A possible explanation
The “far-field” approximation for displacement (f=3Hz, D> 5-6 km):
moment rate
Slip-rate vs M
Active slip area vs M
Rupture kinematics
Rupture dynamics
Dynamic stress drop vs M
Rise-time vs M
The observed correlation between log(PGXt) and Magnitude would imply that
“dynamic stress release” and/or “rise-time”
scale with earthquake size in the very early stage of seismic ruptures
Summary
A high-density, wide-dynamics seismic network is being
installed in southern Italy for “regional” early-warning
applications
The system will implement a real-time eqk location method
based on an evolutionary, probabilistic approach
Early P- and S- signal amplitudes (less the 2 sec from first
arrival) correlate with magnitude (4≤Mw≤7.4) as from the
analysis of the European Strong Motion Data Base
A combination of magnitude estimations obtained by “early
P/S peak amplitudes” and “predominant periods” (Allen &
Kanamori,2003) measured at different stations as a
function of time may significantly improve the accuracy of
the earthquake size estimation in real-time procedures.
The End
Multiple Events
Each time a new pick is available,
the algorithm:
1. Temporarily associates the pick
to the current event
2. Relocates the event
3. Checks the travel-time RMS for
the maximum likelihood
hypocenter
4. If RMS < RMSthresh the pick is
definitively associated,
otherwise a new event is
declared
Vertical
Modulus of
horizontal
components
Strong Motion Data
De Natale et al., BSSA,1987
Kanamori & Rivera, BSSA, 2004
acceleration
velocity
displacement
Log(PGXt) vs Magnitude
Early P-Wave
Early S-Wave
Slide 8
Real-Time Estimation of Earthquake Location
and Magnitude for Seismic Early Warning in
Campania Region, southern Italy
A. Zollo and RISSC-Lab Research Group*
with A.Lomax (A.Lomax Scientific Software)
*
is a joint seismological research group between University of
Naples - Dept of Physics and INGV – Osservatorio Vesuviano
Work Motivation
• Development and testing of a Seismic EarlyWarning System for automated risk mitigation
actions in Campania Region
• Need for robust and reliable real-time estimates
of eqk location and magnitude to be obtained in
an evolving, continually updated form.
• Need to provide with parameter uncertainty
variation with time engineering structural
control
Historical Earthquakes
1980 Irpinia
earthquake,
Ms=6.9
Recent Seismicity
Early Warning
Network
INGV catalogue
(1981-2002)
M2.5
29 sites
Osiris 24-bit Data Logger
6 channels:
3 accelerometers
3 seismometers
(Short Period or
Broad Band)
Real time data analysis
SEW System Peculiarities
Characteristic times:
3-5 s
1.5 - 3.5 s
16 - 18 s
eqk at 4-16 km
depth
Latency
&
computation
60 km
22 - 24 s
80 km
28 - 30 s
100 km
time
To
TP first
TS target
High spatial density :
Station spacing < 15 km
Wide-Dynamics:
Unsaturated signals up to 1 g
Real-Time Earthquake Location
Basic Ideas:
Constraint from “not-yet-triggered” stations
Tracing and crossing Equal Differential
Time (EDT) surfaces
Probabilistic estimation of eqk location vs
time (Evolutionary Approach)
Real-time Evolutionary Location
• When a first station Sn triggers at tn = tnow,
we can already place limits on a pdf
volume that is likely to contain the
hypocenter (Voronoi cell). These limits
are given by conditional EDT surfaces on
which the P travel time to the first
triggering station A is equal to the traveltime to each of the not-yet-triggered
stations.
• As the current time tnow progresses we
gain the additional information that the
not-yet-triggered stations can only trigger
with tl > tnow
• When the second and later stations
trigger, we construct standard, true EDT
surfaces between each pair of the
triggered stations.These EDT surfaces
are stacked with the volume defined by
the not-yet-triggered stations to form the
current hypocentral pdf volume.
Synthetic Examples
Earthquake location probability
Seconds from first trigger
Seconds from earthquake Origin Time
Triggered stations
Earthquake location probability
Real-Time Magnitude Estimate
Basic Ideas
• Use both early P- and S-wave information based
on the high density / wide dynamics of the
network
• Correlate low-pass filtered peak amplitudes with
Magnitude in increasing time windows
• Regression analysis based on the European
strong motion Data-Base (ESD, Ambraseys,
2004)
European Strong Motion Data Base*
* Ambraseys et al. (2004)
• 207 Events with 4≤MW≤7.4 (Kokaeli,1999)
• 376 three-component records
• Epicentral distance ≤ 50 km
• Low-pass filter: 3 Hz
• Magnitude bin: 0.3
Vertical
Measurement of Peak Amplitude
2-sec
H(t)=NS2(t)+EW2(t)
Horizontal
1-sec
3Hz low-pass filtered acceleration at station Bagnoli
(1980, Irpinia Eqk, Ms=6.9) – Epicentral Distance: 20 Km
Log(displacement)
Log(PGDt) vs Magnitude
P-wave
S-wave
Mean value
2-Weighted
Standard Error
Single data point
magnitude
A possible explanation
The “far-field” approximation for displacement (f=3Hz, D> 5-6 km):
moment rate
Slip-rate vs M
Active slip area vs M
Rupture kinematics
Rupture dynamics
Dynamic stress drop vs M
Rise-time vs M
The observed correlation between log(PGXt) and Magnitude would imply that
“dynamic stress release” and/or “rise-time”
scale with earthquake size in the very early stage of seismic ruptures
Summary
A high-density, wide-dynamics seismic network is being
installed in southern Italy for “regional” early-warning
applications
The system will implement a real-time eqk location method
based on an evolutionary, probabilistic approach
Early P- and S- signal amplitudes (less the 2 sec from first
arrival) correlate with magnitude (4≤Mw≤7.4) as from the
analysis of the European Strong Motion Data Base
A combination of magnitude estimations obtained by “early
P/S peak amplitudes” and “predominant periods” (Allen &
Kanamori,2003) measured at different stations as a
function of time may significantly improve the accuracy of
the earthquake size estimation in real-time procedures.
The End
Multiple Events
Each time a new pick is available,
the algorithm:
1. Temporarily associates the pick
to the current event
2. Relocates the event
3. Checks the travel-time RMS for
the maximum likelihood
hypocenter
4. If RMS < RMSthresh the pick is
definitively associated,
otherwise a new event is
declared
Vertical
Modulus of
horizontal
components
Strong Motion Data
De Natale et al., BSSA,1987
Kanamori & Rivera, BSSA, 2004
acceleration
velocity
displacement
Log(PGXt) vs Magnitude
Early P-Wave
Early S-Wave
Slide 9
Real-Time Estimation of Earthquake Location
and Magnitude for Seismic Early Warning in
Campania Region, southern Italy
A. Zollo and RISSC-Lab Research Group*
with A.Lomax (A.Lomax Scientific Software)
*
is a joint seismological research group between University of
Naples - Dept of Physics and INGV – Osservatorio Vesuviano
Work Motivation
• Development and testing of a Seismic EarlyWarning System for automated risk mitigation
actions in Campania Region
• Need for robust and reliable real-time estimates
of eqk location and magnitude to be obtained in
an evolving, continually updated form.
• Need to provide with parameter uncertainty
variation with time engineering structural
control
Historical Earthquakes
1980 Irpinia
earthquake,
Ms=6.9
Recent Seismicity
Early Warning
Network
INGV catalogue
(1981-2002)
M2.5
29 sites
Osiris 24-bit Data Logger
6 channels:
3 accelerometers
3 seismometers
(Short Period or
Broad Band)
Real time data analysis
SEW System Peculiarities
Characteristic times:
3-5 s
1.5 - 3.5 s
16 - 18 s
eqk at 4-16 km
depth
Latency
&
computation
60 km
22 - 24 s
80 km
28 - 30 s
100 km
time
To
TP first
TS target
High spatial density :
Station spacing < 15 km
Wide-Dynamics:
Unsaturated signals up to 1 g
Real-Time Earthquake Location
Basic Ideas:
Constraint from “not-yet-triggered” stations
Tracing and crossing Equal Differential
Time (EDT) surfaces
Probabilistic estimation of eqk location vs
time (Evolutionary Approach)
Real-time Evolutionary Location
• When a first station Sn triggers at tn = tnow,
we can already place limits on a pdf
volume that is likely to contain the
hypocenter (Voronoi cell). These limits
are given by conditional EDT surfaces on
which the P travel time to the first
triggering station A is equal to the traveltime to each of the not-yet-triggered
stations.
• As the current time tnow progresses we
gain the additional information that the
not-yet-triggered stations can only trigger
with tl > tnow
• When the second and later stations
trigger, we construct standard, true EDT
surfaces between each pair of the
triggered stations.These EDT surfaces
are stacked with the volume defined by
the not-yet-triggered stations to form the
current hypocentral pdf volume.
Synthetic Examples
Earthquake location probability
Seconds from first trigger
Seconds from earthquake Origin Time
Triggered stations
Earthquake location probability
Real-Time Magnitude Estimate
Basic Ideas
• Use both early P- and S-wave information based
on the high density / wide dynamics of the
network
• Correlate low-pass filtered peak amplitudes with
Magnitude in increasing time windows
• Regression analysis based on the European
strong motion Data-Base (ESD, Ambraseys,
2004)
European Strong Motion Data Base*
* Ambraseys et al. (2004)
• 207 Events with 4≤MW≤7.4 (Kokaeli,1999)
• 376 three-component records
• Epicentral distance ≤ 50 km
• Low-pass filter: 3 Hz
• Magnitude bin: 0.3
Vertical
Measurement of Peak Amplitude
2-sec
H(t)=NS2(t)+EW2(t)
Horizontal
1-sec
3Hz low-pass filtered acceleration at station Bagnoli
(1980, Irpinia Eqk, Ms=6.9) – Epicentral Distance: 20 Km
Log(displacement)
Log(PGDt) vs Magnitude
P-wave
S-wave
Mean value
2-Weighted
Standard Error
Single data point
magnitude
A possible explanation
The “far-field” approximation for displacement (f=3Hz, D> 5-6 km):
moment rate
Slip-rate vs M
Active slip area vs M
Rupture kinematics
Rupture dynamics
Dynamic stress drop vs M
Rise-time vs M
The observed correlation between log(PGXt) and Magnitude would imply that
“dynamic stress release” and/or “rise-time”
scale with earthquake size in the very early stage of seismic ruptures
Summary
A high-density, wide-dynamics seismic network is being
installed in southern Italy for “regional” early-warning
applications
The system will implement a real-time eqk location method
based on an evolutionary, probabilistic approach
Early P- and S- signal amplitudes (less the 2 sec from first
arrival) correlate with magnitude (4≤Mw≤7.4) as from the
analysis of the European Strong Motion Data Base
A combination of magnitude estimations obtained by “early
P/S peak amplitudes” and “predominant periods” (Allen &
Kanamori,2003) measured at different stations as a
function of time may significantly improve the accuracy of
the earthquake size estimation in real-time procedures.
The End
Multiple Events
Each time a new pick is available,
the algorithm:
1. Temporarily associates the pick
to the current event
2. Relocates the event
3. Checks the travel-time RMS for
the maximum likelihood
hypocenter
4. If RMS < RMSthresh the pick is
definitively associated,
otherwise a new event is
declared
Vertical
Modulus of
horizontal
components
Strong Motion Data
De Natale et al., BSSA,1987
Kanamori & Rivera, BSSA, 2004
acceleration
velocity
displacement
Log(PGXt) vs Magnitude
Early P-Wave
Early S-Wave
Slide 10
Real-Time Estimation of Earthquake Location
and Magnitude for Seismic Early Warning in
Campania Region, southern Italy
A. Zollo and RISSC-Lab Research Group*
with A.Lomax (A.Lomax Scientific Software)
*
is a joint seismological research group between University of
Naples - Dept of Physics and INGV – Osservatorio Vesuviano
Work Motivation
• Development and testing of a Seismic EarlyWarning System for automated risk mitigation
actions in Campania Region
• Need for robust and reliable real-time estimates
of eqk location and magnitude to be obtained in
an evolving, continually updated form.
• Need to provide with parameter uncertainty
variation with time engineering structural
control
Historical Earthquakes
1980 Irpinia
earthquake,
Ms=6.9
Recent Seismicity
Early Warning
Network
INGV catalogue
(1981-2002)
M2.5
29 sites
Osiris 24-bit Data Logger
6 channels:
3 accelerometers
3 seismometers
(Short Period or
Broad Band)
Real time data analysis
SEW System Peculiarities
Characteristic times:
3-5 s
1.5 - 3.5 s
16 - 18 s
eqk at 4-16 km
depth
Latency
&
computation
60 km
22 - 24 s
80 km
28 - 30 s
100 km
time
To
TP first
TS target
High spatial density :
Station spacing < 15 km
Wide-Dynamics:
Unsaturated signals up to 1 g
Real-Time Earthquake Location
Basic Ideas:
Constraint from “not-yet-triggered” stations
Tracing and crossing Equal Differential
Time (EDT) surfaces
Probabilistic estimation of eqk location vs
time (Evolutionary Approach)
Real-time Evolutionary Location
• When a first station Sn triggers at tn = tnow,
we can already place limits on a pdf
volume that is likely to contain the
hypocenter (Voronoi cell). These limits
are given by conditional EDT surfaces on
which the P travel time to the first
triggering station A is equal to the traveltime to each of the not-yet-triggered
stations.
• As the current time tnow progresses we
gain the additional information that the
not-yet-triggered stations can only trigger
with tl > tnow
• When the second and later stations
trigger, we construct standard, true EDT
surfaces between each pair of the
triggered stations.These EDT surfaces
are stacked with the volume defined by
the not-yet-triggered stations to form the
current hypocentral pdf volume.
Synthetic Examples
Earthquake location probability
Seconds from first trigger
Seconds from earthquake Origin Time
Triggered stations
Earthquake location probability
Real-Time Magnitude Estimate
Basic Ideas
• Use both early P- and S-wave information based
on the high density / wide dynamics of the
network
• Correlate low-pass filtered peak amplitudes with
Magnitude in increasing time windows
• Regression analysis based on the European
strong motion Data-Base (ESD, Ambraseys,
2004)
European Strong Motion Data Base*
* Ambraseys et al. (2004)
• 207 Events with 4≤MW≤7.4 (Kokaeli,1999)
• 376 three-component records
• Epicentral distance ≤ 50 km
• Low-pass filter: 3 Hz
• Magnitude bin: 0.3
Vertical
Measurement of Peak Amplitude
2-sec
H(t)=NS2(t)+EW2(t)
Horizontal
1-sec
3Hz low-pass filtered acceleration at station Bagnoli
(1980, Irpinia Eqk, Ms=6.9) – Epicentral Distance: 20 Km
Log(displacement)
Log(PGDt) vs Magnitude
P-wave
S-wave
Mean value
2-Weighted
Standard Error
Single data point
magnitude
A possible explanation
The “far-field” approximation for displacement (f=3Hz, D> 5-6 km):
moment rate
Slip-rate vs M
Active slip area vs M
Rupture kinematics
Rupture dynamics
Dynamic stress drop vs M
Rise-time vs M
The observed correlation between log(PGXt) and Magnitude would imply that
“dynamic stress release” and/or “rise-time”
scale with earthquake size in the very early stage of seismic ruptures
Summary
A high-density, wide-dynamics seismic network is being
installed in southern Italy for “regional” early-warning
applications
The system will implement a real-time eqk location method
based on an evolutionary, probabilistic approach
Early P- and S- signal amplitudes (less the 2 sec from first
arrival) correlate with magnitude (4≤Mw≤7.4) as from the
analysis of the European Strong Motion Data Base
A combination of magnitude estimations obtained by “early
P/S peak amplitudes” and “predominant periods” (Allen &
Kanamori,2003) measured at different stations as a
function of time may significantly improve the accuracy of
the earthquake size estimation in real-time procedures.
The End
Multiple Events
Each time a new pick is available,
the algorithm:
1. Temporarily associates the pick
to the current event
2. Relocates the event
3. Checks the travel-time RMS for
the maximum likelihood
hypocenter
4. If RMS < RMSthresh the pick is
definitively associated,
otherwise a new event is
declared
Vertical
Modulus of
horizontal
components
Strong Motion Data
De Natale et al., BSSA,1987
Kanamori & Rivera, BSSA, 2004
acceleration
velocity
displacement
Log(PGXt) vs Magnitude
Early P-Wave
Early S-Wave
Slide 11
Real-Time Estimation of Earthquake Location
and Magnitude for Seismic Early Warning in
Campania Region, southern Italy
A. Zollo and RISSC-Lab Research Group*
with A.Lomax (A.Lomax Scientific Software)
*
is a joint seismological research group between University of
Naples - Dept of Physics and INGV – Osservatorio Vesuviano
Work Motivation
• Development and testing of a Seismic EarlyWarning System for automated risk mitigation
actions in Campania Region
• Need for robust and reliable real-time estimates
of eqk location and magnitude to be obtained in
an evolving, continually updated form.
• Need to provide with parameter uncertainty
variation with time engineering structural
control
Historical Earthquakes
1980 Irpinia
earthquake,
Ms=6.9
Recent Seismicity
Early Warning
Network
INGV catalogue
(1981-2002)
M2.5
29 sites
Osiris 24-bit Data Logger
6 channels:
3 accelerometers
3 seismometers
(Short Period or
Broad Band)
Real time data analysis
SEW System Peculiarities
Characteristic times:
3-5 s
1.5 - 3.5 s
16 - 18 s
eqk at 4-16 km
depth
Latency
&
computation
60 km
22 - 24 s
80 km
28 - 30 s
100 km
time
To
TP first
TS target
High spatial density :
Station spacing < 15 km
Wide-Dynamics:
Unsaturated signals up to 1 g
Real-Time Earthquake Location
Basic Ideas:
Constraint from “not-yet-triggered” stations
Tracing and crossing Equal Differential
Time (EDT) surfaces
Probabilistic estimation of eqk location vs
time (Evolutionary Approach)
Real-time Evolutionary Location
• When a first station Sn triggers at tn = tnow,
we can already place limits on a pdf
volume that is likely to contain the
hypocenter (Voronoi cell). These limits
are given by conditional EDT surfaces on
which the P travel time to the first
triggering station A is equal to the traveltime to each of the not-yet-triggered
stations.
• As the current time tnow progresses we
gain the additional information that the
not-yet-triggered stations can only trigger
with tl > tnow
• When the second and later stations
trigger, we construct standard, true EDT
surfaces between each pair of the
triggered stations.These EDT surfaces
are stacked with the volume defined by
the not-yet-triggered stations to form the
current hypocentral pdf volume.
Synthetic Examples
Earthquake location probability
Seconds from first trigger
Seconds from earthquake Origin Time
Triggered stations
Earthquake location probability
Real-Time Magnitude Estimate
Basic Ideas
• Use both early P- and S-wave information based
on the high density / wide dynamics of the
network
• Correlate low-pass filtered peak amplitudes with
Magnitude in increasing time windows
• Regression analysis based on the European
strong motion Data-Base (ESD, Ambraseys,
2004)
European Strong Motion Data Base*
* Ambraseys et al. (2004)
• 207 Events with 4≤MW≤7.4 (Kokaeli,1999)
• 376 three-component records
• Epicentral distance ≤ 50 km
• Low-pass filter: 3 Hz
• Magnitude bin: 0.3
Vertical
Measurement of Peak Amplitude
2-sec
H(t)=NS2(t)+EW2(t)
Horizontal
1-sec
3Hz low-pass filtered acceleration at station Bagnoli
(1980, Irpinia Eqk, Ms=6.9) – Epicentral Distance: 20 Km
Log(displacement)
Log(PGDt) vs Magnitude
P-wave
S-wave
Mean value
2-Weighted
Standard Error
Single data point
magnitude
A possible explanation
The “far-field” approximation for displacement (f=3Hz, D> 5-6 km):
moment rate
Slip-rate vs M
Active slip area vs M
Rupture kinematics
Rupture dynamics
Dynamic stress drop vs M
Rise-time vs M
The observed correlation between log(PGXt) and Magnitude would imply that
“dynamic stress release” and/or “rise-time”
scale with earthquake size in the very early stage of seismic ruptures
Summary
A high-density, wide-dynamics seismic network is being
installed in southern Italy for “regional” early-warning
applications
The system will implement a real-time eqk location method
based on an evolutionary, probabilistic approach
Early P- and S- signal amplitudes (less the 2 sec from first
arrival) correlate with magnitude (4≤Mw≤7.4) as from the
analysis of the European Strong Motion Data Base
A combination of magnitude estimations obtained by “early
P/S peak amplitudes” and “predominant periods” (Allen &
Kanamori,2003) measured at different stations as a
function of time may significantly improve the accuracy of
the earthquake size estimation in real-time procedures.
The End
Multiple Events
Each time a new pick is available,
the algorithm:
1. Temporarily associates the pick
to the current event
2. Relocates the event
3. Checks the travel-time RMS for
the maximum likelihood
hypocenter
4. If RMS < RMSthresh the pick is
definitively associated,
otherwise a new event is
declared
Vertical
Modulus of
horizontal
components
Strong Motion Data
De Natale et al., BSSA,1987
Kanamori & Rivera, BSSA, 2004
acceleration
velocity
displacement
Log(PGXt) vs Magnitude
Early P-Wave
Early S-Wave
Slide 12
Real-Time Estimation of Earthquake Location
and Magnitude for Seismic Early Warning in
Campania Region, southern Italy
A. Zollo and RISSC-Lab Research Group*
with A.Lomax (A.Lomax Scientific Software)
*
is a joint seismological research group between University of
Naples - Dept of Physics and INGV – Osservatorio Vesuviano
Work Motivation
• Development and testing of a Seismic EarlyWarning System for automated risk mitigation
actions in Campania Region
• Need for robust and reliable real-time estimates
of eqk location and magnitude to be obtained in
an evolving, continually updated form.
• Need to provide with parameter uncertainty
variation with time engineering structural
control
Historical Earthquakes
1980 Irpinia
earthquake,
Ms=6.9
Recent Seismicity
Early Warning
Network
INGV catalogue
(1981-2002)
M2.5
29 sites
Osiris 24-bit Data Logger
6 channels:
3 accelerometers
3 seismometers
(Short Period or
Broad Band)
Real time data analysis
SEW System Peculiarities
Characteristic times:
3-5 s
1.5 - 3.5 s
16 - 18 s
eqk at 4-16 km
depth
Latency
&
computation
60 km
22 - 24 s
80 km
28 - 30 s
100 km
time
To
TP first
TS target
High spatial density :
Station spacing < 15 km
Wide-Dynamics:
Unsaturated signals up to 1 g
Real-Time Earthquake Location
Basic Ideas:
Constraint from “not-yet-triggered” stations
Tracing and crossing Equal Differential
Time (EDT) surfaces
Probabilistic estimation of eqk location vs
time (Evolutionary Approach)
Real-time Evolutionary Location
• When a first station Sn triggers at tn = tnow,
we can already place limits on a pdf
volume that is likely to contain the
hypocenter (Voronoi cell). These limits
are given by conditional EDT surfaces on
which the P travel time to the first
triggering station A is equal to the traveltime to each of the not-yet-triggered
stations.
• As the current time tnow progresses we
gain the additional information that the
not-yet-triggered stations can only trigger
with tl > tnow
• When the second and later stations
trigger, we construct standard, true EDT
surfaces between each pair of the
triggered stations.These EDT surfaces
are stacked with the volume defined by
the not-yet-triggered stations to form the
current hypocentral pdf volume.
Synthetic Examples
Earthquake location probability
Seconds from first trigger
Seconds from earthquake Origin Time
Triggered stations
Earthquake location probability
Real-Time Magnitude Estimate
Basic Ideas
• Use both early P- and S-wave information based
on the high density / wide dynamics of the
network
• Correlate low-pass filtered peak amplitudes with
Magnitude in increasing time windows
• Regression analysis based on the European
strong motion Data-Base (ESD, Ambraseys,
2004)
European Strong Motion Data Base*
* Ambraseys et al. (2004)
• 207 Events with 4≤MW≤7.4 (Kokaeli,1999)
• 376 three-component records
• Epicentral distance ≤ 50 km
• Low-pass filter: 3 Hz
• Magnitude bin: 0.3
Vertical
Measurement of Peak Amplitude
2-sec
H(t)=NS2(t)+EW2(t)
Horizontal
1-sec
3Hz low-pass filtered acceleration at station Bagnoli
(1980, Irpinia Eqk, Ms=6.9) – Epicentral Distance: 20 Km
Log(displacement)
Log(PGDt) vs Magnitude
P-wave
S-wave
Mean value
2-Weighted
Standard Error
Single data point
magnitude
A possible explanation
The “far-field” approximation for displacement (f=3Hz, D> 5-6 km):
moment rate
Slip-rate vs M
Active slip area vs M
Rupture kinematics
Rupture dynamics
Dynamic stress drop vs M
Rise-time vs M
The observed correlation between log(PGXt) and Magnitude would imply that
“dynamic stress release” and/or “rise-time”
scale with earthquake size in the very early stage of seismic ruptures
Summary
A high-density, wide-dynamics seismic network is being
installed in southern Italy for “regional” early-warning
applications
The system will implement a real-time eqk location method
based on an evolutionary, probabilistic approach
Early P- and S- signal amplitudes (less the 2 sec from first
arrival) correlate with magnitude (4≤Mw≤7.4) as from the
analysis of the European Strong Motion Data Base
A combination of magnitude estimations obtained by “early
P/S peak amplitudes” and “predominant periods” (Allen &
Kanamori,2003) measured at different stations as a
function of time may significantly improve the accuracy of
the earthquake size estimation in real-time procedures.
The End
Multiple Events
Each time a new pick is available,
the algorithm:
1. Temporarily associates the pick
to the current event
2. Relocates the event
3. Checks the travel-time RMS for
the maximum likelihood
hypocenter
4. If RMS < RMSthresh the pick is
definitively associated,
otherwise a new event is
declared
Vertical
Modulus of
horizontal
components
Strong Motion Data
De Natale et al., BSSA,1987
Kanamori & Rivera, BSSA, 2004
acceleration
velocity
displacement
Log(PGXt) vs Magnitude
Early P-Wave
Early S-Wave
Slide 13
Real-Time Estimation of Earthquake Location
and Magnitude for Seismic Early Warning in
Campania Region, southern Italy
A. Zollo and RISSC-Lab Research Group*
with A.Lomax (A.Lomax Scientific Software)
*
is a joint seismological research group between University of
Naples - Dept of Physics and INGV – Osservatorio Vesuviano
Work Motivation
• Development and testing of a Seismic EarlyWarning System for automated risk mitigation
actions in Campania Region
• Need for robust and reliable real-time estimates
of eqk location and magnitude to be obtained in
an evolving, continually updated form.
• Need to provide with parameter uncertainty
variation with time engineering structural
control
Historical Earthquakes
1980 Irpinia
earthquake,
Ms=6.9
Recent Seismicity
Early Warning
Network
INGV catalogue
(1981-2002)
M2.5
29 sites
Osiris 24-bit Data Logger
6 channels:
3 accelerometers
3 seismometers
(Short Period or
Broad Band)
Real time data analysis
SEW System Peculiarities
Characteristic times:
3-5 s
1.5 - 3.5 s
16 - 18 s
eqk at 4-16 km
depth
Latency
&
computation
60 km
22 - 24 s
80 km
28 - 30 s
100 km
time
To
TP first
TS target
High spatial density :
Station spacing < 15 km
Wide-Dynamics:
Unsaturated signals up to 1 g
Real-Time Earthquake Location
Basic Ideas:
Constraint from “not-yet-triggered” stations
Tracing and crossing Equal Differential
Time (EDT) surfaces
Probabilistic estimation of eqk location vs
time (Evolutionary Approach)
Real-time Evolutionary Location
• When a first station Sn triggers at tn = tnow,
we can already place limits on a pdf
volume that is likely to contain the
hypocenter (Voronoi cell). These limits
are given by conditional EDT surfaces on
which the P travel time to the first
triggering station A is equal to the traveltime to each of the not-yet-triggered
stations.
• As the current time tnow progresses we
gain the additional information that the
not-yet-triggered stations can only trigger
with tl > tnow
• When the second and later stations
trigger, we construct standard, true EDT
surfaces between each pair of the
triggered stations.These EDT surfaces
are stacked with the volume defined by
the not-yet-triggered stations to form the
current hypocentral pdf volume.
Synthetic Examples
Earthquake location probability
Seconds from first trigger
Seconds from earthquake Origin Time
Triggered stations
Earthquake location probability
Real-Time Magnitude Estimate
Basic Ideas
• Use both early P- and S-wave information based
on the high density / wide dynamics of the
network
• Correlate low-pass filtered peak amplitudes with
Magnitude in increasing time windows
• Regression analysis based on the European
strong motion Data-Base (ESD, Ambraseys,
2004)
European Strong Motion Data Base*
* Ambraseys et al. (2004)
• 207 Events with 4≤MW≤7.4 (Kokaeli,1999)
• 376 three-component records
• Epicentral distance ≤ 50 km
• Low-pass filter: 3 Hz
• Magnitude bin: 0.3
Vertical
Measurement of Peak Amplitude
2-sec
H(t)=NS2(t)+EW2(t)
Horizontal
1-sec
3Hz low-pass filtered acceleration at station Bagnoli
(1980, Irpinia Eqk, Ms=6.9) – Epicentral Distance: 20 Km
Log(displacement)
Log(PGDt) vs Magnitude
P-wave
S-wave
Mean value
2-Weighted
Standard Error
Single data point
magnitude
A possible explanation
The “far-field” approximation for displacement (f=3Hz, D> 5-6 km):
moment rate
Slip-rate vs M
Active slip area vs M
Rupture kinematics
Rupture dynamics
Dynamic stress drop vs M
Rise-time vs M
The observed correlation between log(PGXt) and Magnitude would imply that
“dynamic stress release” and/or “rise-time”
scale with earthquake size in the very early stage of seismic ruptures
Summary
A high-density, wide-dynamics seismic network is being
installed in southern Italy for “regional” early-warning
applications
The system will implement a real-time eqk location method
based on an evolutionary, probabilistic approach
Early P- and S- signal amplitudes (less the 2 sec from first
arrival) correlate with magnitude (4≤Mw≤7.4) as from the
analysis of the European Strong Motion Data Base
A combination of magnitude estimations obtained by “early
P/S peak amplitudes” and “predominant periods” (Allen &
Kanamori,2003) measured at different stations as a
function of time may significantly improve the accuracy of
the earthquake size estimation in real-time procedures.
The End
Multiple Events
Each time a new pick is available,
the algorithm:
1. Temporarily associates the pick
to the current event
2. Relocates the event
3. Checks the travel-time RMS for
the maximum likelihood
hypocenter
4. If RMS < RMSthresh the pick is
definitively associated,
otherwise a new event is
declared
Vertical
Modulus of
horizontal
components
Strong Motion Data
De Natale et al., BSSA,1987
Kanamori & Rivera, BSSA, 2004
acceleration
velocity
displacement
Log(PGXt) vs Magnitude
Early P-Wave
Early S-Wave
Slide 14
Real-Time Estimation of Earthquake Location
and Magnitude for Seismic Early Warning in
Campania Region, southern Italy
A. Zollo and RISSC-Lab Research Group*
with A.Lomax (A.Lomax Scientific Software)
*
is a joint seismological research group between University of
Naples - Dept of Physics and INGV – Osservatorio Vesuviano
Work Motivation
• Development and testing of a Seismic EarlyWarning System for automated risk mitigation
actions in Campania Region
• Need for robust and reliable real-time estimates
of eqk location and magnitude to be obtained in
an evolving, continually updated form.
• Need to provide with parameter uncertainty
variation with time engineering structural
control
Historical Earthquakes
1980 Irpinia
earthquake,
Ms=6.9
Recent Seismicity
Early Warning
Network
INGV catalogue
(1981-2002)
M2.5
29 sites
Osiris 24-bit Data Logger
6 channels:
3 accelerometers
3 seismometers
(Short Period or
Broad Band)
Real time data analysis
SEW System Peculiarities
Characteristic times:
3-5 s
1.5 - 3.5 s
16 - 18 s
eqk at 4-16 km
depth
Latency
&
computation
60 km
22 - 24 s
80 km
28 - 30 s
100 km
time
To
TP first
TS target
High spatial density :
Station spacing < 15 km
Wide-Dynamics:
Unsaturated signals up to 1 g
Real-Time Earthquake Location
Basic Ideas:
Constraint from “not-yet-triggered” stations
Tracing and crossing Equal Differential
Time (EDT) surfaces
Probabilistic estimation of eqk location vs
time (Evolutionary Approach)
Real-time Evolutionary Location
• When a first station Sn triggers at tn = tnow,
we can already place limits on a pdf
volume that is likely to contain the
hypocenter (Voronoi cell). These limits
are given by conditional EDT surfaces on
which the P travel time to the first
triggering station A is equal to the traveltime to each of the not-yet-triggered
stations.
• As the current time tnow progresses we
gain the additional information that the
not-yet-triggered stations can only trigger
with tl > tnow
• When the second and later stations
trigger, we construct standard, true EDT
surfaces between each pair of the
triggered stations.These EDT surfaces
are stacked with the volume defined by
the not-yet-triggered stations to form the
current hypocentral pdf volume.
Synthetic Examples
Earthquake location probability
Seconds from first trigger
Seconds from earthquake Origin Time
Triggered stations
Earthquake location probability
Real-Time Magnitude Estimate
Basic Ideas
• Use both early P- and S-wave information based
on the high density / wide dynamics of the
network
• Correlate low-pass filtered peak amplitudes with
Magnitude in increasing time windows
• Regression analysis based on the European
strong motion Data-Base (ESD, Ambraseys,
2004)
European Strong Motion Data Base*
* Ambraseys et al. (2004)
• 207 Events with 4≤MW≤7.4 (Kokaeli,1999)
• 376 three-component records
• Epicentral distance ≤ 50 km
• Low-pass filter: 3 Hz
• Magnitude bin: 0.3
Vertical
Measurement of Peak Amplitude
2-sec
H(t)=NS2(t)+EW2(t)
Horizontal
1-sec
3Hz low-pass filtered acceleration at station Bagnoli
(1980, Irpinia Eqk, Ms=6.9) – Epicentral Distance: 20 Km
Log(displacement)
Log(PGDt) vs Magnitude
P-wave
S-wave
Mean value
2-Weighted
Standard Error
Single data point
magnitude
A possible explanation
The “far-field” approximation for displacement (f=3Hz, D> 5-6 km):
moment rate
Slip-rate vs M
Active slip area vs M
Rupture kinematics
Rupture dynamics
Dynamic stress drop vs M
Rise-time vs M
The observed correlation between log(PGXt) and Magnitude would imply that
“dynamic stress release” and/or “rise-time”
scale with earthquake size in the very early stage of seismic ruptures
Summary
A high-density, wide-dynamics seismic network is being
installed in southern Italy for “regional” early-warning
applications
The system will implement a real-time eqk location method
based on an evolutionary, probabilistic approach
Early P- and S- signal amplitudes (less the 2 sec from first
arrival) correlate with magnitude (4≤Mw≤7.4) as from the
analysis of the European Strong Motion Data Base
A combination of magnitude estimations obtained by “early
P/S peak amplitudes” and “predominant periods” (Allen &
Kanamori,2003) measured at different stations as a
function of time may significantly improve the accuracy of
the earthquake size estimation in real-time procedures.
The End
Multiple Events
Each time a new pick is available,
the algorithm:
1. Temporarily associates the pick
to the current event
2. Relocates the event
3. Checks the travel-time RMS for
the maximum likelihood
hypocenter
4. If RMS < RMSthresh the pick is
definitively associated,
otherwise a new event is
declared
Vertical
Modulus of
horizontal
components
Strong Motion Data
De Natale et al., BSSA,1987
Kanamori & Rivera, BSSA, 2004
acceleration
velocity
displacement
Log(PGXt) vs Magnitude
Early P-Wave
Early S-Wave
Slide 15
Real-Time Estimation of Earthquake Location
and Magnitude for Seismic Early Warning in
Campania Region, southern Italy
A. Zollo and RISSC-Lab Research Group*
with A.Lomax (A.Lomax Scientific Software)
*
is a joint seismological research group between University of
Naples - Dept of Physics and INGV – Osservatorio Vesuviano
Work Motivation
• Development and testing of a Seismic EarlyWarning System for automated risk mitigation
actions in Campania Region
• Need for robust and reliable real-time estimates
of eqk location and magnitude to be obtained in
an evolving, continually updated form.
• Need to provide with parameter uncertainty
variation with time engineering structural
control
Historical Earthquakes
1980 Irpinia
earthquake,
Ms=6.9
Recent Seismicity
Early Warning
Network
INGV catalogue
(1981-2002)
M2.5
29 sites
Osiris 24-bit Data Logger
6 channels:
3 accelerometers
3 seismometers
(Short Period or
Broad Band)
Real time data analysis
SEW System Peculiarities
Characteristic times:
3-5 s
1.5 - 3.5 s
16 - 18 s
eqk at 4-16 km
depth
Latency
&
computation
60 km
22 - 24 s
80 km
28 - 30 s
100 km
time
To
TP first
TS target
High spatial density :
Station spacing < 15 km
Wide-Dynamics:
Unsaturated signals up to 1 g
Real-Time Earthquake Location
Basic Ideas:
Constraint from “not-yet-triggered” stations
Tracing and crossing Equal Differential
Time (EDT) surfaces
Probabilistic estimation of eqk location vs
time (Evolutionary Approach)
Real-time Evolutionary Location
• When a first station Sn triggers at tn = tnow,
we can already place limits on a pdf
volume that is likely to contain the
hypocenter (Voronoi cell). These limits
are given by conditional EDT surfaces on
which the P travel time to the first
triggering station A is equal to the traveltime to each of the not-yet-triggered
stations.
• As the current time tnow progresses we
gain the additional information that the
not-yet-triggered stations can only trigger
with tl > tnow
• When the second and later stations
trigger, we construct standard, true EDT
surfaces between each pair of the
triggered stations.These EDT surfaces
are stacked with the volume defined by
the not-yet-triggered stations to form the
current hypocentral pdf volume.
Synthetic Examples
Earthquake location probability
Seconds from first trigger
Seconds from earthquake Origin Time
Triggered stations
Earthquake location probability
Real-Time Magnitude Estimate
Basic Ideas
• Use both early P- and S-wave information based
on the high density / wide dynamics of the
network
• Correlate low-pass filtered peak amplitudes with
Magnitude in increasing time windows
• Regression analysis based on the European
strong motion Data-Base (ESD, Ambraseys,
2004)
European Strong Motion Data Base*
* Ambraseys et al. (2004)
• 207 Events with 4≤MW≤7.4 (Kokaeli,1999)
• 376 three-component records
• Epicentral distance ≤ 50 km
• Low-pass filter: 3 Hz
• Magnitude bin: 0.3
Vertical
Measurement of Peak Amplitude
2-sec
H(t)=NS2(t)+EW2(t)
Horizontal
1-sec
3Hz low-pass filtered acceleration at station Bagnoli
(1980, Irpinia Eqk, Ms=6.9) – Epicentral Distance: 20 Km
Log(displacement)
Log(PGDt) vs Magnitude
P-wave
S-wave
Mean value
2-Weighted
Standard Error
Single data point
magnitude
A possible explanation
The “far-field” approximation for displacement (f=3Hz, D> 5-6 km):
moment rate
Slip-rate vs M
Active slip area vs M
Rupture kinematics
Rupture dynamics
Dynamic stress drop vs M
Rise-time vs M
The observed correlation between log(PGXt) and Magnitude would imply that
“dynamic stress release” and/or “rise-time”
scale with earthquake size in the very early stage of seismic ruptures
Summary
A high-density, wide-dynamics seismic network is being
installed in southern Italy for “regional” early-warning
applications
The system will implement a real-time eqk location method
based on an evolutionary, probabilistic approach
Early P- and S- signal amplitudes (less the 2 sec from first
arrival) correlate with magnitude (4≤Mw≤7.4) as from the
analysis of the European Strong Motion Data Base
A combination of magnitude estimations obtained by “early
P/S peak amplitudes” and “predominant periods” (Allen &
Kanamori,2003) measured at different stations as a
function of time may significantly improve the accuracy of
the earthquake size estimation in real-time procedures.
The End
Multiple Events
Each time a new pick is available,
the algorithm:
1. Temporarily associates the pick
to the current event
2. Relocates the event
3. Checks the travel-time RMS for
the maximum likelihood
hypocenter
4. If RMS < RMSthresh the pick is
definitively associated,
otherwise a new event is
declared
Vertical
Modulus of
horizontal
components
Strong Motion Data
De Natale et al., BSSA,1987
Kanamori & Rivera, BSSA, 2004
acceleration
velocity
displacement
Log(PGXt) vs Magnitude
Early P-Wave
Early S-Wave
Slide 16
Real-Time Estimation of Earthquake Location
and Magnitude for Seismic Early Warning in
Campania Region, southern Italy
A. Zollo and RISSC-Lab Research Group*
with A.Lomax (A.Lomax Scientific Software)
*
is a joint seismological research group between University of
Naples - Dept of Physics and INGV – Osservatorio Vesuviano
Work Motivation
• Development and testing of a Seismic EarlyWarning System for automated risk mitigation
actions in Campania Region
• Need for robust and reliable real-time estimates
of eqk location and magnitude to be obtained in
an evolving, continually updated form.
• Need to provide with parameter uncertainty
variation with time engineering structural
control
Historical Earthquakes
1980 Irpinia
earthquake,
Ms=6.9
Recent Seismicity
Early Warning
Network
INGV catalogue
(1981-2002)
M2.5
29 sites
Osiris 24-bit Data Logger
6 channels:
3 accelerometers
3 seismometers
(Short Period or
Broad Band)
Real time data analysis
SEW System Peculiarities
Characteristic times:
3-5 s
1.5 - 3.5 s
16 - 18 s
eqk at 4-16 km
depth
Latency
&
computation
60 km
22 - 24 s
80 km
28 - 30 s
100 km
time
To
TP first
TS target
High spatial density :
Station spacing < 15 km
Wide-Dynamics:
Unsaturated signals up to 1 g
Real-Time Earthquake Location
Basic Ideas:
Constraint from “not-yet-triggered” stations
Tracing and crossing Equal Differential
Time (EDT) surfaces
Probabilistic estimation of eqk location vs
time (Evolutionary Approach)
Real-time Evolutionary Location
• When a first station Sn triggers at tn = tnow,
we can already place limits on a pdf
volume that is likely to contain the
hypocenter (Voronoi cell). These limits
are given by conditional EDT surfaces on
which the P travel time to the first
triggering station A is equal to the traveltime to each of the not-yet-triggered
stations.
• As the current time tnow progresses we
gain the additional information that the
not-yet-triggered stations can only trigger
with tl > tnow
• When the second and later stations
trigger, we construct standard, true EDT
surfaces between each pair of the
triggered stations.These EDT surfaces
are stacked with the volume defined by
the not-yet-triggered stations to form the
current hypocentral pdf volume.
Synthetic Examples
Earthquake location probability
Seconds from first trigger
Seconds from earthquake Origin Time
Triggered stations
Earthquake location probability
Real-Time Magnitude Estimate
Basic Ideas
• Use both early P- and S-wave information based
on the high density / wide dynamics of the
network
• Correlate low-pass filtered peak amplitudes with
Magnitude in increasing time windows
• Regression analysis based on the European
strong motion Data-Base (ESD, Ambraseys,
2004)
European Strong Motion Data Base*
* Ambraseys et al. (2004)
• 207 Events with 4≤MW≤7.4 (Kokaeli,1999)
• 376 three-component records
• Epicentral distance ≤ 50 km
• Low-pass filter: 3 Hz
• Magnitude bin: 0.3
Vertical
Measurement of Peak Amplitude
2-sec
H(t)=NS2(t)+EW2(t)
Horizontal
1-sec
3Hz low-pass filtered acceleration at station Bagnoli
(1980, Irpinia Eqk, Ms=6.9) – Epicentral Distance: 20 Km
Log(displacement)
Log(PGDt) vs Magnitude
P-wave
S-wave
Mean value
2-Weighted
Standard Error
Single data point
magnitude
A possible explanation
The “far-field” approximation for displacement (f=3Hz, D> 5-6 km):
moment rate
Slip-rate vs M
Active slip area vs M
Rupture kinematics
Rupture dynamics
Dynamic stress drop vs M
Rise-time vs M
The observed correlation between log(PGXt) and Magnitude would imply that
“dynamic stress release” and/or “rise-time”
scale with earthquake size in the very early stage of seismic ruptures
Summary
A high-density, wide-dynamics seismic network is being
installed in southern Italy for “regional” early-warning
applications
The system will implement a real-time eqk location method
based on an evolutionary, probabilistic approach
Early P- and S- signal amplitudes (less the 2 sec from first
arrival) correlate with magnitude (4≤Mw≤7.4) as from the
analysis of the European Strong Motion Data Base
A combination of magnitude estimations obtained by “early
P/S peak amplitudes” and “predominant periods” (Allen &
Kanamori,2003) measured at different stations as a
function of time may significantly improve the accuracy of
the earthquake size estimation in real-time procedures.
The End
Multiple Events
Each time a new pick is available,
the algorithm:
1. Temporarily associates the pick
to the current event
2. Relocates the event
3. Checks the travel-time RMS for
the maximum likelihood
hypocenter
4. If RMS < RMSthresh the pick is
definitively associated,
otherwise a new event is
declared
Vertical
Modulus of
horizontal
components
Strong Motion Data
De Natale et al., BSSA,1987
Kanamori & Rivera, BSSA, 2004
acceleration
velocity
displacement
Log(PGXt) vs Magnitude
Early P-Wave
Early S-Wave
Slide 17
Real-Time Estimation of Earthquake Location
and Magnitude for Seismic Early Warning in
Campania Region, southern Italy
A. Zollo and RISSC-Lab Research Group*
with A.Lomax (A.Lomax Scientific Software)
*
is a joint seismological research group between University of
Naples - Dept of Physics and INGV – Osservatorio Vesuviano
Work Motivation
• Development and testing of a Seismic EarlyWarning System for automated risk mitigation
actions in Campania Region
• Need for robust and reliable real-time estimates
of eqk location and magnitude to be obtained in
an evolving, continually updated form.
• Need to provide with parameter uncertainty
variation with time engineering structural
control
Historical Earthquakes
1980 Irpinia
earthquake,
Ms=6.9
Recent Seismicity
Early Warning
Network
INGV catalogue
(1981-2002)
M2.5
29 sites
Osiris 24-bit Data Logger
6 channels:
3 accelerometers
3 seismometers
(Short Period or
Broad Band)
Real time data analysis
SEW System Peculiarities
Characteristic times:
3-5 s
1.5 - 3.5 s
16 - 18 s
eqk at 4-16 km
depth
Latency
&
computation
60 km
22 - 24 s
80 km
28 - 30 s
100 km
time
To
TP first
TS target
High spatial density :
Station spacing < 15 km
Wide-Dynamics:
Unsaturated signals up to 1 g
Real-Time Earthquake Location
Basic Ideas:
Constraint from “not-yet-triggered” stations
Tracing and crossing Equal Differential
Time (EDT) surfaces
Probabilistic estimation of eqk location vs
time (Evolutionary Approach)
Real-time Evolutionary Location
• When a first station Sn triggers at tn = tnow,
we can already place limits on a pdf
volume that is likely to contain the
hypocenter (Voronoi cell). These limits
are given by conditional EDT surfaces on
which the P travel time to the first
triggering station A is equal to the traveltime to each of the not-yet-triggered
stations.
• As the current time tnow progresses we
gain the additional information that the
not-yet-triggered stations can only trigger
with tl > tnow
• When the second and later stations
trigger, we construct standard, true EDT
surfaces between each pair of the
triggered stations.These EDT surfaces
are stacked with the volume defined by
the not-yet-triggered stations to form the
current hypocentral pdf volume.
Synthetic Examples
Earthquake location probability
Seconds from first trigger
Seconds from earthquake Origin Time
Triggered stations
Earthquake location probability
Real-Time Magnitude Estimate
Basic Ideas
• Use both early P- and S-wave information based
on the high density / wide dynamics of the
network
• Correlate low-pass filtered peak amplitudes with
Magnitude in increasing time windows
• Regression analysis based on the European
strong motion Data-Base (ESD, Ambraseys,
2004)
European Strong Motion Data Base*
* Ambraseys et al. (2004)
• 207 Events with 4≤MW≤7.4 (Kokaeli,1999)
• 376 three-component records
• Epicentral distance ≤ 50 km
• Low-pass filter: 3 Hz
• Magnitude bin: 0.3
Vertical
Measurement of Peak Amplitude
2-sec
H(t)=NS2(t)+EW2(t)
Horizontal
1-sec
3Hz low-pass filtered acceleration at station Bagnoli
(1980, Irpinia Eqk, Ms=6.9) – Epicentral Distance: 20 Km
Log(displacement)
Log(PGDt) vs Magnitude
P-wave
S-wave
Mean value
2-Weighted
Standard Error
Single data point
magnitude
A possible explanation
The “far-field” approximation for displacement (f=3Hz, D> 5-6 km):
moment rate
Slip-rate vs M
Active slip area vs M
Rupture kinematics
Rupture dynamics
Dynamic stress drop vs M
Rise-time vs M
The observed correlation between log(PGXt) and Magnitude would imply that
“dynamic stress release” and/or “rise-time”
scale with earthquake size in the very early stage of seismic ruptures
Summary
A high-density, wide-dynamics seismic network is being
installed in southern Italy for “regional” early-warning
applications
The system will implement a real-time eqk location method
based on an evolutionary, probabilistic approach
Early P- and S- signal amplitudes (less the 2 sec from first
arrival) correlate with magnitude (4≤Mw≤7.4) as from the
analysis of the European Strong Motion Data Base
A combination of magnitude estimations obtained by “early
P/S peak amplitudes” and “predominant periods” (Allen &
Kanamori,2003) measured at different stations as a
function of time may significantly improve the accuracy of
the earthquake size estimation in real-time procedures.
The End
Multiple Events
Each time a new pick is available,
the algorithm:
1. Temporarily associates the pick
to the current event
2. Relocates the event
3. Checks the travel-time RMS for
the maximum likelihood
hypocenter
4. If RMS < RMSthresh the pick is
definitively associated,
otherwise a new event is
declared
Vertical
Modulus of
horizontal
components
Strong Motion Data
De Natale et al., BSSA,1987
Kanamori & Rivera, BSSA, 2004
acceleration
velocity
displacement
Log(PGXt) vs Magnitude
Early P-Wave
Early S-Wave
Slide 18
Real-Time Estimation of Earthquake Location
and Magnitude for Seismic Early Warning in
Campania Region, southern Italy
A. Zollo and RISSC-Lab Research Group*
with A.Lomax (A.Lomax Scientific Software)
*
is a joint seismological research group between University of
Naples - Dept of Physics and INGV – Osservatorio Vesuviano
Work Motivation
• Development and testing of a Seismic EarlyWarning System for automated risk mitigation
actions in Campania Region
• Need for robust and reliable real-time estimates
of eqk location and magnitude to be obtained in
an evolving, continually updated form.
• Need to provide with parameter uncertainty
variation with time engineering structural
control
Historical Earthquakes
1980 Irpinia
earthquake,
Ms=6.9
Recent Seismicity
Early Warning
Network
INGV catalogue
(1981-2002)
M2.5
29 sites
Osiris 24-bit Data Logger
6 channels:
3 accelerometers
3 seismometers
(Short Period or
Broad Band)
Real time data analysis
SEW System Peculiarities
Characteristic times:
3-5 s
1.5 - 3.5 s
16 - 18 s
eqk at 4-16 km
depth
Latency
&
computation
60 km
22 - 24 s
80 km
28 - 30 s
100 km
time
To
TP first
TS target
High spatial density :
Station spacing < 15 km
Wide-Dynamics:
Unsaturated signals up to 1 g
Real-Time Earthquake Location
Basic Ideas:
Constraint from “not-yet-triggered” stations
Tracing and crossing Equal Differential
Time (EDT) surfaces
Probabilistic estimation of eqk location vs
time (Evolutionary Approach)
Real-time Evolutionary Location
• When a first station Sn triggers at tn = tnow,
we can already place limits on a pdf
volume that is likely to contain the
hypocenter (Voronoi cell). These limits
are given by conditional EDT surfaces on
which the P travel time to the first
triggering station A is equal to the traveltime to each of the not-yet-triggered
stations.
• As the current time tnow progresses we
gain the additional information that the
not-yet-triggered stations can only trigger
with tl > tnow
• When the second and later stations
trigger, we construct standard, true EDT
surfaces between each pair of the
triggered stations.These EDT surfaces
are stacked with the volume defined by
the not-yet-triggered stations to form the
current hypocentral pdf volume.
Synthetic Examples
Earthquake location probability
Seconds from first trigger
Seconds from earthquake Origin Time
Triggered stations
Earthquake location probability
Real-Time Magnitude Estimate
Basic Ideas
• Use both early P- and S-wave information based
on the high density / wide dynamics of the
network
• Correlate low-pass filtered peak amplitudes with
Magnitude in increasing time windows
• Regression analysis based on the European
strong motion Data-Base (ESD, Ambraseys,
2004)
European Strong Motion Data Base*
* Ambraseys et al. (2004)
• 207 Events with 4≤MW≤7.4 (Kokaeli,1999)
• 376 three-component records
• Epicentral distance ≤ 50 km
• Low-pass filter: 3 Hz
• Magnitude bin: 0.3
Vertical
Measurement of Peak Amplitude
2-sec
H(t)=NS2(t)+EW2(t)
Horizontal
1-sec
3Hz low-pass filtered acceleration at station Bagnoli
(1980, Irpinia Eqk, Ms=6.9) – Epicentral Distance: 20 Km
Log(displacement)
Log(PGDt) vs Magnitude
P-wave
S-wave
Mean value
2-Weighted
Standard Error
Single data point
magnitude
A possible explanation
The “far-field” approximation for displacement (f=3Hz, D> 5-6 km):
moment rate
Slip-rate vs M
Active slip area vs M
Rupture kinematics
Rupture dynamics
Dynamic stress drop vs M
Rise-time vs M
The observed correlation between log(PGXt) and Magnitude would imply that
“dynamic stress release” and/or “rise-time”
scale with earthquake size in the very early stage of seismic ruptures
Summary
A high-density, wide-dynamics seismic network is being
installed in southern Italy for “regional” early-warning
applications
The system will implement a real-time eqk location method
based on an evolutionary, probabilistic approach
Early P- and S- signal amplitudes (less the 2 sec from first
arrival) correlate with magnitude (4≤Mw≤7.4) as from the
analysis of the European Strong Motion Data Base
A combination of magnitude estimations obtained by “early
P/S peak amplitudes” and “predominant periods” (Allen &
Kanamori,2003) measured at different stations as a
function of time may significantly improve the accuracy of
the earthquake size estimation in real-time procedures.
The End
Multiple Events
Each time a new pick is available,
the algorithm:
1. Temporarily associates the pick
to the current event
2. Relocates the event
3. Checks the travel-time RMS for
the maximum likelihood
hypocenter
4. If RMS < RMSthresh the pick is
definitively associated,
otherwise a new event is
declared
Vertical
Modulus of
horizontal
components
Strong Motion Data
De Natale et al., BSSA,1987
Kanamori & Rivera, BSSA, 2004
acceleration
velocity
displacement
Log(PGXt) vs Magnitude
Early P-Wave
Early S-Wave
Slide 19
Real-Time Estimation of Earthquake Location
and Magnitude for Seismic Early Warning in
Campania Region, southern Italy
A. Zollo and RISSC-Lab Research Group*
with A.Lomax (A.Lomax Scientific Software)
*
is a joint seismological research group between University of
Naples - Dept of Physics and INGV – Osservatorio Vesuviano
Work Motivation
• Development and testing of a Seismic EarlyWarning System for automated risk mitigation
actions in Campania Region
• Need for robust and reliable real-time estimates
of eqk location and magnitude to be obtained in
an evolving, continually updated form.
• Need to provide with parameter uncertainty
variation with time engineering structural
control
Historical Earthquakes
1980 Irpinia
earthquake,
Ms=6.9
Recent Seismicity
Early Warning
Network
INGV catalogue
(1981-2002)
M2.5
29 sites
Osiris 24-bit Data Logger
6 channels:
3 accelerometers
3 seismometers
(Short Period or
Broad Band)
Real time data analysis
SEW System Peculiarities
Characteristic times:
3-5 s
1.5 - 3.5 s
16 - 18 s
eqk at 4-16 km
depth
Latency
&
computation
60 km
22 - 24 s
80 km
28 - 30 s
100 km
time
To
TP first
TS target
High spatial density :
Station spacing < 15 km
Wide-Dynamics:
Unsaturated signals up to 1 g
Real-Time Earthquake Location
Basic Ideas:
Constraint from “not-yet-triggered” stations
Tracing and crossing Equal Differential
Time (EDT) surfaces
Probabilistic estimation of eqk location vs
time (Evolutionary Approach)
Real-time Evolutionary Location
• When a first station Sn triggers at tn = tnow,
we can already place limits on a pdf
volume that is likely to contain the
hypocenter (Voronoi cell). These limits
are given by conditional EDT surfaces on
which the P travel time to the first
triggering station A is equal to the traveltime to each of the not-yet-triggered
stations.
• As the current time tnow progresses we
gain the additional information that the
not-yet-triggered stations can only trigger
with tl > tnow
• When the second and later stations
trigger, we construct standard, true EDT
surfaces between each pair of the
triggered stations.These EDT surfaces
are stacked with the volume defined by
the not-yet-triggered stations to form the
current hypocentral pdf volume.
Synthetic Examples
Earthquake location probability
Seconds from first trigger
Seconds from earthquake Origin Time
Triggered stations
Earthquake location probability
Real-Time Magnitude Estimate
Basic Ideas
• Use both early P- and S-wave information based
on the high density / wide dynamics of the
network
• Correlate low-pass filtered peak amplitudes with
Magnitude in increasing time windows
• Regression analysis based on the European
strong motion Data-Base (ESD, Ambraseys,
2004)
European Strong Motion Data Base*
* Ambraseys et al. (2004)
• 207 Events with 4≤MW≤7.4 (Kokaeli,1999)
• 376 three-component records
• Epicentral distance ≤ 50 km
• Low-pass filter: 3 Hz
• Magnitude bin: 0.3
Vertical
Measurement of Peak Amplitude
2-sec
H(t)=NS2(t)+EW2(t)
Horizontal
1-sec
3Hz low-pass filtered acceleration at station Bagnoli
(1980, Irpinia Eqk, Ms=6.9) – Epicentral Distance: 20 Km
Log(displacement)
Log(PGDt) vs Magnitude
P-wave
S-wave
Mean value
2-Weighted
Standard Error
Single data point
magnitude
A possible explanation
The “far-field” approximation for displacement (f=3Hz, D> 5-6 km):
moment rate
Slip-rate vs M
Active slip area vs M
Rupture kinematics
Rupture dynamics
Dynamic stress drop vs M
Rise-time vs M
The observed correlation between log(PGXt) and Magnitude would imply that
“dynamic stress release” and/or “rise-time”
scale with earthquake size in the very early stage of seismic ruptures
Summary
A high-density, wide-dynamics seismic network is being
installed in southern Italy for “regional” early-warning
applications
The system will implement a real-time eqk location method
based on an evolutionary, probabilistic approach
Early P- and S- signal amplitudes (less the 2 sec from first
arrival) correlate with magnitude (4≤Mw≤7.4) as from the
analysis of the European Strong Motion Data Base
A combination of magnitude estimations obtained by “early
P/S peak amplitudes” and “predominant periods” (Allen &
Kanamori,2003) measured at different stations as a
function of time may significantly improve the accuracy of
the earthquake size estimation in real-time procedures.
The End
Multiple Events
Each time a new pick is available,
the algorithm:
1. Temporarily associates the pick
to the current event
2. Relocates the event
3. Checks the travel-time RMS for
the maximum likelihood
hypocenter
4. If RMS < RMSthresh the pick is
definitively associated,
otherwise a new event is
declared
Vertical
Modulus of
horizontal
components
Strong Motion Data
De Natale et al., BSSA,1987
Kanamori & Rivera, BSSA, 2004
acceleration
velocity
displacement
Log(PGXt) vs Magnitude
Early P-Wave
Early S-Wave