Moment tensor inversion at Pyhasalmi ore mine: accuracy test using explosions Daniela Kühn (NORSAR) V.
Download ReportTranscript Moment tensor inversion at Pyhasalmi ore mine: accuracy test using explosions Daniela Kühn (NORSAR) V.
Slide 1
Moment tensor inversion at
Pyhasalmi ore mine:
accuracy test using explosions
Daniela Kühn (NORSAR)
V. Vavrycuk (Academy of Sciences of the CR)
AIM 2nd annual meeting, 29-30 Sept 2011, Prague, Czech Republic
Pyhäsalmi ore mine, Finland
• microseismic monitoring:
since January 2003
safety of the underground
personnel
optimisation of mining
process
Introduction
Waveform
modelling
Moment
tensor
inversion
• network:
12 1-C geophones
+ 6 3-C geophones (ISS)
3-D geometry
sampling rate: < 3000 Hz
Summary
owned by Inmet Mining Co.
• events:
1500 events /months
(including blasting)
-2 < Mw < 1.5
Inconsistent polarities of P-wave first onset
Moment tensor inversion:
• homogeneous velocity model (as for locations)
• amplitude inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling
Complexity of velocity model
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling: 2D section
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
• E3D: viscoelastic 3-D FD code (Larsen and Schultz, 1995)
• strong interaction with mining cavities: reflection,
scattering, conversion
620 m
• healing of wavefronts
Waveform modelling
synthetic seismograms
- complex waveforms
Introduction
- strong coda
- complex secondary arrivals
Waveform
modelling
- scattering effects stronger
on amplitudes than travel
times,
since
size
of
heterogeneities
(cavities,
access tunnels) same order or
smaller than wavelengths
Moment
tensor
inversion
Summary
arrival times computed by
Eikonal solver still fit
(wavefronts heal quickly
after passing a cavitiy)
observed seismograms
Comparison 1-D/3-D
Imaginary
network!
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Comparison 1-D/3-D
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Influence of proximity to cavity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Real mine
network!
Source depth → ray path
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Ray path → onset polarity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Moment tensor inversion
Amplitude picking
direct wave scattered wave
Introduction
Waveform
modelling
Moment
tensor
inversion
first maximum amplitude = amplitude of the direct wave
direct wave
scattered wave
Summary
energy diffracted
around cavity
?
first maximum amplitude is not always the amplitude of the direct wave
Amplitude vs. waveform inversion
Amplitude inversion:
• homogeneous model of the medium
Introduction
• Green’s functions calculated using ray theory
• inversion of P-wave amplitudes (20-30 amplitudes)
Waveform
modelling
• frequencies: 250-500 Hz
cannot take into account distortion of rays on focal sphere
Moment
tensor
inversion
misinterpretation of amplitudes: which one is the direct wave?
Waveform inversion:
Summary
• 3-D heterogeneous model of the medium
• Green’s functions calculated using FD code
• inversion of full waveforms (15-20 waveforms)
• frequencies (at the moment): 25-100 Hz
inversion is performed in frequency domain
in principle same inversion algorithm as for amplitude inversion, but run
repeatedly for every frequency band (0.5 Hz steps)
Selected explosions: relocation
expl 1
expl 1
expl 3
Introduction
expl 5
Waveform
modelling
expl 3
Moment
tensor
inversion
expl 5
expl 1
Summary
expl 3
expl 5
Explosions (coords in m):
1) x=8306E y=2312N z=-1238
→ 26 m shift
3) x=8218E y=2192N z=-1352
→ 68 m shift
5) x=8214E y=2168N z=-1356
→ 59 m shift
Examples: good fit
Examples: amplitude misfit
Examples: phase misfit
Inversion results
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
explosion 1
explosion 3
explosion 5
Length of time windows
GF duration [10 ms]
Data duration [10 ms]
explosion 1
ISO percentage
GF duration [10 ms]
ISO percentage
GF duration [10 ms]
ISO percentage
Data duration [10 ms]
explosion 3
Data duration [10 ms]
explosion 5
nearly all solutions have high isotropic percentage
best solutions near diagonal
(length of GF time window = length of data time window)
best solutions P –wave + S-wave onset
Explosion 1
Amplitude inversion
Waveform inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
DC = 7%
CLVD = -14%
ISO = 79%
Summary: seismicity in mines
structural model in mines is very complex
Introduction
large and abrupt changes in velocity at
cavities
model varies in time
Waveform
modelling
Moment
tensor
inversion
Summary
earthquake source is complex (single
forces, non-DC components)
small changes in source position lead to
large changes in ray propagation, rays can
be strongly curved
radiated wave field is complex
(reflected, converted, scattered waves, head
waves)
Summary: waveform inversion
In general:
• complex Green’s functions can be calculated by 3-D FD codes (accurate model needed!);
Introduction
• sensitive to time shifts due to mislocation or due to inaccurate velocity model
• frequency band of inverted waves can be easily controlled => stability analysis
Waveform
modelling
In particular:
Moment
tensor
inversion
• good network configuration => focal sphere nicely covered
• inversion algorithm:
• optimal with same window length for Green’s functions and data
Summary
• optimal with simultaneous inversion of P- and S-wave, but excluding S-wave coda
• yields high isotropic percentage, higher than amplitude inversion, almost
independently of window length
promising, but computationally demanding (especially the computation of Green’s
functions with sufficiently small grid point distances)
will be performed for selected events, not whole database
Thank you
for your
attention!
http://commons.wikimedia.org/wiki/File:Preikestolen_Norge.jpg
Slide 2
Moment tensor inversion at
Pyhasalmi ore mine:
accuracy test using explosions
Daniela Kühn (NORSAR)
V. Vavrycuk (Academy of Sciences of the CR)
AIM 2nd annual meeting, 29-30 Sept 2011, Prague, Czech Republic
Pyhäsalmi ore mine, Finland
• microseismic monitoring:
since January 2003
safety of the underground
personnel
optimisation of mining
process
Introduction
Waveform
modelling
Moment
tensor
inversion
• network:
12 1-C geophones
+ 6 3-C geophones (ISS)
3-D geometry
sampling rate: < 3000 Hz
Summary
owned by Inmet Mining Co.
• events:
1500 events /months
(including blasting)
-2 < Mw < 1.5
Inconsistent polarities of P-wave first onset
Moment tensor inversion:
• homogeneous velocity model (as for locations)
• amplitude inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling
Complexity of velocity model
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling: 2D section
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
• E3D: viscoelastic 3-D FD code (Larsen and Schultz, 1995)
• strong interaction with mining cavities: reflection,
scattering, conversion
620 m
• healing of wavefronts
Waveform modelling
synthetic seismograms
- complex waveforms
Introduction
- strong coda
- complex secondary arrivals
Waveform
modelling
- scattering effects stronger
on amplitudes than travel
times,
since
size
of
heterogeneities
(cavities,
access tunnels) same order or
smaller than wavelengths
Moment
tensor
inversion
Summary
arrival times computed by
Eikonal solver still fit
(wavefronts heal quickly
after passing a cavitiy)
observed seismograms
Comparison 1-D/3-D
Imaginary
network!
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Comparison 1-D/3-D
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Influence of proximity to cavity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Real mine
network!
Source depth → ray path
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Ray path → onset polarity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Moment tensor inversion
Amplitude picking
direct wave scattered wave
Introduction
Waveform
modelling
Moment
tensor
inversion
first maximum amplitude = amplitude of the direct wave
direct wave
scattered wave
Summary
energy diffracted
around cavity
?
first maximum amplitude is not always the amplitude of the direct wave
Amplitude vs. waveform inversion
Amplitude inversion:
• homogeneous model of the medium
Introduction
• Green’s functions calculated using ray theory
• inversion of P-wave amplitudes (20-30 amplitudes)
Waveform
modelling
• frequencies: 250-500 Hz
cannot take into account distortion of rays on focal sphere
Moment
tensor
inversion
misinterpretation of amplitudes: which one is the direct wave?
Waveform inversion:
Summary
• 3-D heterogeneous model of the medium
• Green’s functions calculated using FD code
• inversion of full waveforms (15-20 waveforms)
• frequencies (at the moment): 25-100 Hz
inversion is performed in frequency domain
in principle same inversion algorithm as for amplitude inversion, but run
repeatedly for every frequency band (0.5 Hz steps)
Selected explosions: relocation
expl 1
expl 1
expl 3
Introduction
expl 5
Waveform
modelling
expl 3
Moment
tensor
inversion
expl 5
expl 1
Summary
expl 3
expl 5
Explosions (coords in m):
1) x=8306E y=2312N z=-1238
→ 26 m shift
3) x=8218E y=2192N z=-1352
→ 68 m shift
5) x=8214E y=2168N z=-1356
→ 59 m shift
Examples: good fit
Examples: amplitude misfit
Examples: phase misfit
Inversion results
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
explosion 1
explosion 3
explosion 5
Length of time windows
GF duration [10 ms]
Data duration [10 ms]
explosion 1
ISO percentage
GF duration [10 ms]
ISO percentage
GF duration [10 ms]
ISO percentage
Data duration [10 ms]
explosion 3
Data duration [10 ms]
explosion 5
nearly all solutions have high isotropic percentage
best solutions near diagonal
(length of GF time window = length of data time window)
best solutions P –wave + S-wave onset
Explosion 1
Amplitude inversion
Waveform inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
DC = 7%
CLVD = -14%
ISO = 79%
Summary: seismicity in mines
structural model in mines is very complex
Introduction
large and abrupt changes in velocity at
cavities
model varies in time
Waveform
modelling
Moment
tensor
inversion
Summary
earthquake source is complex (single
forces, non-DC components)
small changes in source position lead to
large changes in ray propagation, rays can
be strongly curved
radiated wave field is complex
(reflected, converted, scattered waves, head
waves)
Summary: waveform inversion
In general:
• complex Green’s functions can be calculated by 3-D FD codes (accurate model needed!);
Introduction
• sensitive to time shifts due to mislocation or due to inaccurate velocity model
• frequency band of inverted waves can be easily controlled => stability analysis
Waveform
modelling
In particular:
Moment
tensor
inversion
• good network configuration => focal sphere nicely covered
• inversion algorithm:
• optimal with same window length for Green’s functions and data
Summary
• optimal with simultaneous inversion of P- and S-wave, but excluding S-wave coda
• yields high isotropic percentage, higher than amplitude inversion, almost
independently of window length
promising, but computationally demanding (especially the computation of Green’s
functions with sufficiently small grid point distances)
will be performed for selected events, not whole database
Thank you
for your
attention!
http://commons.wikimedia.org/wiki/File:Preikestolen_Norge.jpg
Slide 3
Moment tensor inversion at
Pyhasalmi ore mine:
accuracy test using explosions
Daniela Kühn (NORSAR)
V. Vavrycuk (Academy of Sciences of the CR)
AIM 2nd annual meeting, 29-30 Sept 2011, Prague, Czech Republic
Pyhäsalmi ore mine, Finland
• microseismic monitoring:
since January 2003
safety of the underground
personnel
optimisation of mining
process
Introduction
Waveform
modelling
Moment
tensor
inversion
• network:
12 1-C geophones
+ 6 3-C geophones (ISS)
3-D geometry
sampling rate: < 3000 Hz
Summary
owned by Inmet Mining Co.
• events:
1500 events /months
(including blasting)
-2 < Mw < 1.5
Inconsistent polarities of P-wave first onset
Moment tensor inversion:
• homogeneous velocity model (as for locations)
• amplitude inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling
Complexity of velocity model
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling: 2D section
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
• E3D: viscoelastic 3-D FD code (Larsen and Schultz, 1995)
• strong interaction with mining cavities: reflection,
scattering, conversion
620 m
• healing of wavefronts
Waveform modelling
synthetic seismograms
- complex waveforms
Introduction
- strong coda
- complex secondary arrivals
Waveform
modelling
- scattering effects stronger
on amplitudes than travel
times,
since
size
of
heterogeneities
(cavities,
access tunnels) same order or
smaller than wavelengths
Moment
tensor
inversion
Summary
arrival times computed by
Eikonal solver still fit
(wavefronts heal quickly
after passing a cavitiy)
observed seismograms
Comparison 1-D/3-D
Imaginary
network!
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Comparison 1-D/3-D
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Influence of proximity to cavity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Real mine
network!
Source depth → ray path
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Ray path → onset polarity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Moment tensor inversion
Amplitude picking
direct wave scattered wave
Introduction
Waveform
modelling
Moment
tensor
inversion
first maximum amplitude = amplitude of the direct wave
direct wave
scattered wave
Summary
energy diffracted
around cavity
?
first maximum amplitude is not always the amplitude of the direct wave
Amplitude vs. waveform inversion
Amplitude inversion:
• homogeneous model of the medium
Introduction
• Green’s functions calculated using ray theory
• inversion of P-wave amplitudes (20-30 amplitudes)
Waveform
modelling
• frequencies: 250-500 Hz
cannot take into account distortion of rays on focal sphere
Moment
tensor
inversion
misinterpretation of amplitudes: which one is the direct wave?
Waveform inversion:
Summary
• 3-D heterogeneous model of the medium
• Green’s functions calculated using FD code
• inversion of full waveforms (15-20 waveforms)
• frequencies (at the moment): 25-100 Hz
inversion is performed in frequency domain
in principle same inversion algorithm as for amplitude inversion, but run
repeatedly for every frequency band (0.5 Hz steps)
Selected explosions: relocation
expl 1
expl 1
expl 3
Introduction
expl 5
Waveform
modelling
expl 3
Moment
tensor
inversion
expl 5
expl 1
Summary
expl 3
expl 5
Explosions (coords in m):
1) x=8306E y=2312N z=-1238
→ 26 m shift
3) x=8218E y=2192N z=-1352
→ 68 m shift
5) x=8214E y=2168N z=-1356
→ 59 m shift
Examples: good fit
Examples: amplitude misfit
Examples: phase misfit
Inversion results
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
explosion 1
explosion 3
explosion 5
Length of time windows
GF duration [10 ms]
Data duration [10 ms]
explosion 1
ISO percentage
GF duration [10 ms]
ISO percentage
GF duration [10 ms]
ISO percentage
Data duration [10 ms]
explosion 3
Data duration [10 ms]
explosion 5
nearly all solutions have high isotropic percentage
best solutions near diagonal
(length of GF time window = length of data time window)
best solutions P –wave + S-wave onset
Explosion 1
Amplitude inversion
Waveform inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
DC = 7%
CLVD = -14%
ISO = 79%
Summary: seismicity in mines
structural model in mines is very complex
Introduction
large and abrupt changes in velocity at
cavities
model varies in time
Waveform
modelling
Moment
tensor
inversion
Summary
earthquake source is complex (single
forces, non-DC components)
small changes in source position lead to
large changes in ray propagation, rays can
be strongly curved
radiated wave field is complex
(reflected, converted, scattered waves, head
waves)
Summary: waveform inversion
In general:
• complex Green’s functions can be calculated by 3-D FD codes (accurate model needed!);
Introduction
• sensitive to time shifts due to mislocation or due to inaccurate velocity model
• frequency band of inverted waves can be easily controlled => stability analysis
Waveform
modelling
In particular:
Moment
tensor
inversion
• good network configuration => focal sphere nicely covered
• inversion algorithm:
• optimal with same window length for Green’s functions and data
Summary
• optimal with simultaneous inversion of P- and S-wave, but excluding S-wave coda
• yields high isotropic percentage, higher than amplitude inversion, almost
independently of window length
promising, but computationally demanding (especially the computation of Green’s
functions with sufficiently small grid point distances)
will be performed for selected events, not whole database
Thank you
for your
attention!
http://commons.wikimedia.org/wiki/File:Preikestolen_Norge.jpg
Slide 4
Moment tensor inversion at
Pyhasalmi ore mine:
accuracy test using explosions
Daniela Kühn (NORSAR)
V. Vavrycuk (Academy of Sciences of the CR)
AIM 2nd annual meeting, 29-30 Sept 2011, Prague, Czech Republic
Pyhäsalmi ore mine, Finland
• microseismic monitoring:
since January 2003
safety of the underground
personnel
optimisation of mining
process
Introduction
Waveform
modelling
Moment
tensor
inversion
• network:
12 1-C geophones
+ 6 3-C geophones (ISS)
3-D geometry
sampling rate: < 3000 Hz
Summary
owned by Inmet Mining Co.
• events:
1500 events /months
(including blasting)
-2 < Mw < 1.5
Inconsistent polarities of P-wave first onset
Moment tensor inversion:
• homogeneous velocity model (as for locations)
• amplitude inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling
Complexity of velocity model
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling: 2D section
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
• E3D: viscoelastic 3-D FD code (Larsen and Schultz, 1995)
• strong interaction with mining cavities: reflection,
scattering, conversion
620 m
• healing of wavefronts
Waveform modelling
synthetic seismograms
- complex waveforms
Introduction
- strong coda
- complex secondary arrivals
Waveform
modelling
- scattering effects stronger
on amplitudes than travel
times,
since
size
of
heterogeneities
(cavities,
access tunnels) same order or
smaller than wavelengths
Moment
tensor
inversion
Summary
arrival times computed by
Eikonal solver still fit
(wavefronts heal quickly
after passing a cavitiy)
observed seismograms
Comparison 1-D/3-D
Imaginary
network!
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Comparison 1-D/3-D
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Influence of proximity to cavity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Real mine
network!
Source depth → ray path
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Ray path → onset polarity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Moment tensor inversion
Amplitude picking
direct wave scattered wave
Introduction
Waveform
modelling
Moment
tensor
inversion
first maximum amplitude = amplitude of the direct wave
direct wave
scattered wave
Summary
energy diffracted
around cavity
?
first maximum amplitude is not always the amplitude of the direct wave
Amplitude vs. waveform inversion
Amplitude inversion:
• homogeneous model of the medium
Introduction
• Green’s functions calculated using ray theory
• inversion of P-wave amplitudes (20-30 amplitudes)
Waveform
modelling
• frequencies: 250-500 Hz
cannot take into account distortion of rays on focal sphere
Moment
tensor
inversion
misinterpretation of amplitudes: which one is the direct wave?
Waveform inversion:
Summary
• 3-D heterogeneous model of the medium
• Green’s functions calculated using FD code
• inversion of full waveforms (15-20 waveforms)
• frequencies (at the moment): 25-100 Hz
inversion is performed in frequency domain
in principle same inversion algorithm as for amplitude inversion, but run
repeatedly for every frequency band (0.5 Hz steps)
Selected explosions: relocation
expl 1
expl 1
expl 3
Introduction
expl 5
Waveform
modelling
expl 3
Moment
tensor
inversion
expl 5
expl 1
Summary
expl 3
expl 5
Explosions (coords in m):
1) x=8306E y=2312N z=-1238
→ 26 m shift
3) x=8218E y=2192N z=-1352
→ 68 m shift
5) x=8214E y=2168N z=-1356
→ 59 m shift
Examples: good fit
Examples: amplitude misfit
Examples: phase misfit
Inversion results
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
explosion 1
explosion 3
explosion 5
Length of time windows
GF duration [10 ms]
Data duration [10 ms]
explosion 1
ISO percentage
GF duration [10 ms]
ISO percentage
GF duration [10 ms]
ISO percentage
Data duration [10 ms]
explosion 3
Data duration [10 ms]
explosion 5
nearly all solutions have high isotropic percentage
best solutions near diagonal
(length of GF time window = length of data time window)
best solutions P –wave + S-wave onset
Explosion 1
Amplitude inversion
Waveform inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
DC = 7%
CLVD = -14%
ISO = 79%
Summary: seismicity in mines
structural model in mines is very complex
Introduction
large and abrupt changes in velocity at
cavities
model varies in time
Waveform
modelling
Moment
tensor
inversion
Summary
earthquake source is complex (single
forces, non-DC components)
small changes in source position lead to
large changes in ray propagation, rays can
be strongly curved
radiated wave field is complex
(reflected, converted, scattered waves, head
waves)
Summary: waveform inversion
In general:
• complex Green’s functions can be calculated by 3-D FD codes (accurate model needed!);
Introduction
• sensitive to time shifts due to mislocation or due to inaccurate velocity model
• frequency band of inverted waves can be easily controlled => stability analysis
Waveform
modelling
In particular:
Moment
tensor
inversion
• good network configuration => focal sphere nicely covered
• inversion algorithm:
• optimal with same window length for Green’s functions and data
Summary
• optimal with simultaneous inversion of P- and S-wave, but excluding S-wave coda
• yields high isotropic percentage, higher than amplitude inversion, almost
independently of window length
promising, but computationally demanding (especially the computation of Green’s
functions with sufficiently small grid point distances)
will be performed for selected events, not whole database
Thank you
for your
attention!
http://commons.wikimedia.org/wiki/File:Preikestolen_Norge.jpg
Slide 5
Moment tensor inversion at
Pyhasalmi ore mine:
accuracy test using explosions
Daniela Kühn (NORSAR)
V. Vavrycuk (Academy of Sciences of the CR)
AIM 2nd annual meeting, 29-30 Sept 2011, Prague, Czech Republic
Pyhäsalmi ore mine, Finland
• microseismic monitoring:
since January 2003
safety of the underground
personnel
optimisation of mining
process
Introduction
Waveform
modelling
Moment
tensor
inversion
• network:
12 1-C geophones
+ 6 3-C geophones (ISS)
3-D geometry
sampling rate: < 3000 Hz
Summary
owned by Inmet Mining Co.
• events:
1500 events /months
(including blasting)
-2 < Mw < 1.5
Inconsistent polarities of P-wave first onset
Moment tensor inversion:
• homogeneous velocity model (as for locations)
• amplitude inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling
Complexity of velocity model
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling: 2D section
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
• E3D: viscoelastic 3-D FD code (Larsen and Schultz, 1995)
• strong interaction with mining cavities: reflection,
scattering, conversion
620 m
• healing of wavefronts
Waveform modelling
synthetic seismograms
- complex waveforms
Introduction
- strong coda
- complex secondary arrivals
Waveform
modelling
- scattering effects stronger
on amplitudes than travel
times,
since
size
of
heterogeneities
(cavities,
access tunnels) same order or
smaller than wavelengths
Moment
tensor
inversion
Summary
arrival times computed by
Eikonal solver still fit
(wavefronts heal quickly
after passing a cavitiy)
observed seismograms
Comparison 1-D/3-D
Imaginary
network!
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Comparison 1-D/3-D
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Influence of proximity to cavity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Real mine
network!
Source depth → ray path
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Ray path → onset polarity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Moment tensor inversion
Amplitude picking
direct wave scattered wave
Introduction
Waveform
modelling
Moment
tensor
inversion
first maximum amplitude = amplitude of the direct wave
direct wave
scattered wave
Summary
energy diffracted
around cavity
?
first maximum amplitude is not always the amplitude of the direct wave
Amplitude vs. waveform inversion
Amplitude inversion:
• homogeneous model of the medium
Introduction
• Green’s functions calculated using ray theory
• inversion of P-wave amplitudes (20-30 amplitudes)
Waveform
modelling
• frequencies: 250-500 Hz
cannot take into account distortion of rays on focal sphere
Moment
tensor
inversion
misinterpretation of amplitudes: which one is the direct wave?
Waveform inversion:
Summary
• 3-D heterogeneous model of the medium
• Green’s functions calculated using FD code
• inversion of full waveforms (15-20 waveforms)
• frequencies (at the moment): 25-100 Hz
inversion is performed in frequency domain
in principle same inversion algorithm as for amplitude inversion, but run
repeatedly for every frequency band (0.5 Hz steps)
Selected explosions: relocation
expl 1
expl 1
expl 3
Introduction
expl 5
Waveform
modelling
expl 3
Moment
tensor
inversion
expl 5
expl 1
Summary
expl 3
expl 5
Explosions (coords in m):
1) x=8306E y=2312N z=-1238
→ 26 m shift
3) x=8218E y=2192N z=-1352
→ 68 m shift
5) x=8214E y=2168N z=-1356
→ 59 m shift
Examples: good fit
Examples: amplitude misfit
Examples: phase misfit
Inversion results
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
explosion 1
explosion 3
explosion 5
Length of time windows
GF duration [10 ms]
Data duration [10 ms]
explosion 1
ISO percentage
GF duration [10 ms]
ISO percentage
GF duration [10 ms]
ISO percentage
Data duration [10 ms]
explosion 3
Data duration [10 ms]
explosion 5
nearly all solutions have high isotropic percentage
best solutions near diagonal
(length of GF time window = length of data time window)
best solutions P –wave + S-wave onset
Explosion 1
Amplitude inversion
Waveform inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
DC = 7%
CLVD = -14%
ISO = 79%
Summary: seismicity in mines
structural model in mines is very complex
Introduction
large and abrupt changes in velocity at
cavities
model varies in time
Waveform
modelling
Moment
tensor
inversion
Summary
earthquake source is complex (single
forces, non-DC components)
small changes in source position lead to
large changes in ray propagation, rays can
be strongly curved
radiated wave field is complex
(reflected, converted, scattered waves, head
waves)
Summary: waveform inversion
In general:
• complex Green’s functions can be calculated by 3-D FD codes (accurate model needed!);
Introduction
• sensitive to time shifts due to mislocation or due to inaccurate velocity model
• frequency band of inverted waves can be easily controlled => stability analysis
Waveform
modelling
In particular:
Moment
tensor
inversion
• good network configuration => focal sphere nicely covered
• inversion algorithm:
• optimal with same window length for Green’s functions and data
Summary
• optimal with simultaneous inversion of P- and S-wave, but excluding S-wave coda
• yields high isotropic percentage, higher than amplitude inversion, almost
independently of window length
promising, but computationally demanding (especially the computation of Green’s
functions with sufficiently small grid point distances)
will be performed for selected events, not whole database
Thank you
for your
attention!
http://commons.wikimedia.org/wiki/File:Preikestolen_Norge.jpg
Slide 6
Moment tensor inversion at
Pyhasalmi ore mine:
accuracy test using explosions
Daniela Kühn (NORSAR)
V. Vavrycuk (Academy of Sciences of the CR)
AIM 2nd annual meeting, 29-30 Sept 2011, Prague, Czech Republic
Pyhäsalmi ore mine, Finland
• microseismic monitoring:
since January 2003
safety of the underground
personnel
optimisation of mining
process
Introduction
Waveform
modelling
Moment
tensor
inversion
• network:
12 1-C geophones
+ 6 3-C geophones (ISS)
3-D geometry
sampling rate: < 3000 Hz
Summary
owned by Inmet Mining Co.
• events:
1500 events /months
(including blasting)
-2 < Mw < 1.5
Inconsistent polarities of P-wave first onset
Moment tensor inversion:
• homogeneous velocity model (as for locations)
• amplitude inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling
Complexity of velocity model
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling: 2D section
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
• E3D: viscoelastic 3-D FD code (Larsen and Schultz, 1995)
• strong interaction with mining cavities: reflection,
scattering, conversion
620 m
• healing of wavefronts
Waveform modelling
synthetic seismograms
- complex waveforms
Introduction
- strong coda
- complex secondary arrivals
Waveform
modelling
- scattering effects stronger
on amplitudes than travel
times,
since
size
of
heterogeneities
(cavities,
access tunnels) same order or
smaller than wavelengths
Moment
tensor
inversion
Summary
arrival times computed by
Eikonal solver still fit
(wavefronts heal quickly
after passing a cavitiy)
observed seismograms
Comparison 1-D/3-D
Imaginary
network!
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Comparison 1-D/3-D
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Influence of proximity to cavity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Real mine
network!
Source depth → ray path
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Ray path → onset polarity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Moment tensor inversion
Amplitude picking
direct wave scattered wave
Introduction
Waveform
modelling
Moment
tensor
inversion
first maximum amplitude = amplitude of the direct wave
direct wave
scattered wave
Summary
energy diffracted
around cavity
?
first maximum amplitude is not always the amplitude of the direct wave
Amplitude vs. waveform inversion
Amplitude inversion:
• homogeneous model of the medium
Introduction
• Green’s functions calculated using ray theory
• inversion of P-wave amplitudes (20-30 amplitudes)
Waveform
modelling
• frequencies: 250-500 Hz
cannot take into account distortion of rays on focal sphere
Moment
tensor
inversion
misinterpretation of amplitudes: which one is the direct wave?
Waveform inversion:
Summary
• 3-D heterogeneous model of the medium
• Green’s functions calculated using FD code
• inversion of full waveforms (15-20 waveforms)
• frequencies (at the moment): 25-100 Hz
inversion is performed in frequency domain
in principle same inversion algorithm as for amplitude inversion, but run
repeatedly for every frequency band (0.5 Hz steps)
Selected explosions: relocation
expl 1
expl 1
expl 3
Introduction
expl 5
Waveform
modelling
expl 3
Moment
tensor
inversion
expl 5
expl 1
Summary
expl 3
expl 5
Explosions (coords in m):
1) x=8306E y=2312N z=-1238
→ 26 m shift
3) x=8218E y=2192N z=-1352
→ 68 m shift
5) x=8214E y=2168N z=-1356
→ 59 m shift
Examples: good fit
Examples: amplitude misfit
Examples: phase misfit
Inversion results
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
explosion 1
explosion 3
explosion 5
Length of time windows
GF duration [10 ms]
Data duration [10 ms]
explosion 1
ISO percentage
GF duration [10 ms]
ISO percentage
GF duration [10 ms]
ISO percentage
Data duration [10 ms]
explosion 3
Data duration [10 ms]
explosion 5
nearly all solutions have high isotropic percentage
best solutions near diagonal
(length of GF time window = length of data time window)
best solutions P –wave + S-wave onset
Explosion 1
Amplitude inversion
Waveform inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
DC = 7%
CLVD = -14%
ISO = 79%
Summary: seismicity in mines
structural model in mines is very complex
Introduction
large and abrupt changes in velocity at
cavities
model varies in time
Waveform
modelling
Moment
tensor
inversion
Summary
earthquake source is complex (single
forces, non-DC components)
small changes in source position lead to
large changes in ray propagation, rays can
be strongly curved
radiated wave field is complex
(reflected, converted, scattered waves, head
waves)
Summary: waveform inversion
In general:
• complex Green’s functions can be calculated by 3-D FD codes (accurate model needed!);
Introduction
• sensitive to time shifts due to mislocation or due to inaccurate velocity model
• frequency band of inverted waves can be easily controlled => stability analysis
Waveform
modelling
In particular:
Moment
tensor
inversion
• good network configuration => focal sphere nicely covered
• inversion algorithm:
• optimal with same window length for Green’s functions and data
Summary
• optimal with simultaneous inversion of P- and S-wave, but excluding S-wave coda
• yields high isotropic percentage, higher than amplitude inversion, almost
independently of window length
promising, but computationally demanding (especially the computation of Green’s
functions with sufficiently small grid point distances)
will be performed for selected events, not whole database
Thank you
for your
attention!
http://commons.wikimedia.org/wiki/File:Preikestolen_Norge.jpg
Slide 7
Moment tensor inversion at
Pyhasalmi ore mine:
accuracy test using explosions
Daniela Kühn (NORSAR)
V. Vavrycuk (Academy of Sciences of the CR)
AIM 2nd annual meeting, 29-30 Sept 2011, Prague, Czech Republic
Pyhäsalmi ore mine, Finland
• microseismic monitoring:
since January 2003
safety of the underground
personnel
optimisation of mining
process
Introduction
Waveform
modelling
Moment
tensor
inversion
• network:
12 1-C geophones
+ 6 3-C geophones (ISS)
3-D geometry
sampling rate: < 3000 Hz
Summary
owned by Inmet Mining Co.
• events:
1500 events /months
(including blasting)
-2 < Mw < 1.5
Inconsistent polarities of P-wave first onset
Moment tensor inversion:
• homogeneous velocity model (as for locations)
• amplitude inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling
Complexity of velocity model
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling: 2D section
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
• E3D: viscoelastic 3-D FD code (Larsen and Schultz, 1995)
• strong interaction with mining cavities: reflection,
scattering, conversion
620 m
• healing of wavefronts
Waveform modelling
synthetic seismograms
- complex waveforms
Introduction
- strong coda
- complex secondary arrivals
Waveform
modelling
- scattering effects stronger
on amplitudes than travel
times,
since
size
of
heterogeneities
(cavities,
access tunnels) same order or
smaller than wavelengths
Moment
tensor
inversion
Summary
arrival times computed by
Eikonal solver still fit
(wavefronts heal quickly
after passing a cavitiy)
observed seismograms
Comparison 1-D/3-D
Imaginary
network!
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Comparison 1-D/3-D
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Influence of proximity to cavity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Real mine
network!
Source depth → ray path
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Ray path → onset polarity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Moment tensor inversion
Amplitude picking
direct wave scattered wave
Introduction
Waveform
modelling
Moment
tensor
inversion
first maximum amplitude = amplitude of the direct wave
direct wave
scattered wave
Summary
energy diffracted
around cavity
?
first maximum amplitude is not always the amplitude of the direct wave
Amplitude vs. waveform inversion
Amplitude inversion:
• homogeneous model of the medium
Introduction
• Green’s functions calculated using ray theory
• inversion of P-wave amplitudes (20-30 amplitudes)
Waveform
modelling
• frequencies: 250-500 Hz
cannot take into account distortion of rays on focal sphere
Moment
tensor
inversion
misinterpretation of amplitudes: which one is the direct wave?
Waveform inversion:
Summary
• 3-D heterogeneous model of the medium
• Green’s functions calculated using FD code
• inversion of full waveforms (15-20 waveforms)
• frequencies (at the moment): 25-100 Hz
inversion is performed in frequency domain
in principle same inversion algorithm as for amplitude inversion, but run
repeatedly for every frequency band (0.5 Hz steps)
Selected explosions: relocation
expl 1
expl 1
expl 3
Introduction
expl 5
Waveform
modelling
expl 3
Moment
tensor
inversion
expl 5
expl 1
Summary
expl 3
expl 5
Explosions (coords in m):
1) x=8306E y=2312N z=-1238
→ 26 m shift
3) x=8218E y=2192N z=-1352
→ 68 m shift
5) x=8214E y=2168N z=-1356
→ 59 m shift
Examples: good fit
Examples: amplitude misfit
Examples: phase misfit
Inversion results
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
explosion 1
explosion 3
explosion 5
Length of time windows
GF duration [10 ms]
Data duration [10 ms]
explosion 1
ISO percentage
GF duration [10 ms]
ISO percentage
GF duration [10 ms]
ISO percentage
Data duration [10 ms]
explosion 3
Data duration [10 ms]
explosion 5
nearly all solutions have high isotropic percentage
best solutions near diagonal
(length of GF time window = length of data time window)
best solutions P –wave + S-wave onset
Explosion 1
Amplitude inversion
Waveform inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
DC = 7%
CLVD = -14%
ISO = 79%
Summary: seismicity in mines
structural model in mines is very complex
Introduction
large and abrupt changes in velocity at
cavities
model varies in time
Waveform
modelling
Moment
tensor
inversion
Summary
earthquake source is complex (single
forces, non-DC components)
small changes in source position lead to
large changes in ray propagation, rays can
be strongly curved
radiated wave field is complex
(reflected, converted, scattered waves, head
waves)
Summary: waveform inversion
In general:
• complex Green’s functions can be calculated by 3-D FD codes (accurate model needed!);
Introduction
• sensitive to time shifts due to mislocation or due to inaccurate velocity model
• frequency band of inverted waves can be easily controlled => stability analysis
Waveform
modelling
In particular:
Moment
tensor
inversion
• good network configuration => focal sphere nicely covered
• inversion algorithm:
• optimal with same window length for Green’s functions and data
Summary
• optimal with simultaneous inversion of P- and S-wave, but excluding S-wave coda
• yields high isotropic percentage, higher than amplitude inversion, almost
independently of window length
promising, but computationally demanding (especially the computation of Green’s
functions with sufficiently small grid point distances)
will be performed for selected events, not whole database
Thank you
for your
attention!
http://commons.wikimedia.org/wiki/File:Preikestolen_Norge.jpg
Slide 8
Moment tensor inversion at
Pyhasalmi ore mine:
accuracy test using explosions
Daniela Kühn (NORSAR)
V. Vavrycuk (Academy of Sciences of the CR)
AIM 2nd annual meeting, 29-30 Sept 2011, Prague, Czech Republic
Pyhäsalmi ore mine, Finland
• microseismic monitoring:
since January 2003
safety of the underground
personnel
optimisation of mining
process
Introduction
Waveform
modelling
Moment
tensor
inversion
• network:
12 1-C geophones
+ 6 3-C geophones (ISS)
3-D geometry
sampling rate: < 3000 Hz
Summary
owned by Inmet Mining Co.
• events:
1500 events /months
(including blasting)
-2 < Mw < 1.5
Inconsistent polarities of P-wave first onset
Moment tensor inversion:
• homogeneous velocity model (as for locations)
• amplitude inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling
Complexity of velocity model
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling: 2D section
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
• E3D: viscoelastic 3-D FD code (Larsen and Schultz, 1995)
• strong interaction with mining cavities: reflection,
scattering, conversion
620 m
• healing of wavefronts
Waveform modelling
synthetic seismograms
- complex waveforms
Introduction
- strong coda
- complex secondary arrivals
Waveform
modelling
- scattering effects stronger
on amplitudes than travel
times,
since
size
of
heterogeneities
(cavities,
access tunnels) same order or
smaller than wavelengths
Moment
tensor
inversion
Summary
arrival times computed by
Eikonal solver still fit
(wavefronts heal quickly
after passing a cavitiy)
observed seismograms
Comparison 1-D/3-D
Imaginary
network!
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Comparison 1-D/3-D
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Influence of proximity to cavity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Real mine
network!
Source depth → ray path
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Ray path → onset polarity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Moment tensor inversion
Amplitude picking
direct wave scattered wave
Introduction
Waveform
modelling
Moment
tensor
inversion
first maximum amplitude = amplitude of the direct wave
direct wave
scattered wave
Summary
energy diffracted
around cavity
?
first maximum amplitude is not always the amplitude of the direct wave
Amplitude vs. waveform inversion
Amplitude inversion:
• homogeneous model of the medium
Introduction
• Green’s functions calculated using ray theory
• inversion of P-wave amplitudes (20-30 amplitudes)
Waveform
modelling
• frequencies: 250-500 Hz
cannot take into account distortion of rays on focal sphere
Moment
tensor
inversion
misinterpretation of amplitudes: which one is the direct wave?
Waveform inversion:
Summary
• 3-D heterogeneous model of the medium
• Green’s functions calculated using FD code
• inversion of full waveforms (15-20 waveforms)
• frequencies (at the moment): 25-100 Hz
inversion is performed in frequency domain
in principle same inversion algorithm as for amplitude inversion, but run
repeatedly for every frequency band (0.5 Hz steps)
Selected explosions: relocation
expl 1
expl 1
expl 3
Introduction
expl 5
Waveform
modelling
expl 3
Moment
tensor
inversion
expl 5
expl 1
Summary
expl 3
expl 5
Explosions (coords in m):
1) x=8306E y=2312N z=-1238
→ 26 m shift
3) x=8218E y=2192N z=-1352
→ 68 m shift
5) x=8214E y=2168N z=-1356
→ 59 m shift
Examples: good fit
Examples: amplitude misfit
Examples: phase misfit
Inversion results
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
explosion 1
explosion 3
explosion 5
Length of time windows
GF duration [10 ms]
Data duration [10 ms]
explosion 1
ISO percentage
GF duration [10 ms]
ISO percentage
GF duration [10 ms]
ISO percentage
Data duration [10 ms]
explosion 3
Data duration [10 ms]
explosion 5
nearly all solutions have high isotropic percentage
best solutions near diagonal
(length of GF time window = length of data time window)
best solutions P –wave + S-wave onset
Explosion 1
Amplitude inversion
Waveform inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
DC = 7%
CLVD = -14%
ISO = 79%
Summary: seismicity in mines
structural model in mines is very complex
Introduction
large and abrupt changes in velocity at
cavities
model varies in time
Waveform
modelling
Moment
tensor
inversion
Summary
earthquake source is complex (single
forces, non-DC components)
small changes in source position lead to
large changes in ray propagation, rays can
be strongly curved
radiated wave field is complex
(reflected, converted, scattered waves, head
waves)
Summary: waveform inversion
In general:
• complex Green’s functions can be calculated by 3-D FD codes (accurate model needed!);
Introduction
• sensitive to time shifts due to mislocation or due to inaccurate velocity model
• frequency band of inverted waves can be easily controlled => stability analysis
Waveform
modelling
In particular:
Moment
tensor
inversion
• good network configuration => focal sphere nicely covered
• inversion algorithm:
• optimal with same window length for Green’s functions and data
Summary
• optimal with simultaneous inversion of P- and S-wave, but excluding S-wave coda
• yields high isotropic percentage, higher than amplitude inversion, almost
independently of window length
promising, but computationally demanding (especially the computation of Green’s
functions with sufficiently small grid point distances)
will be performed for selected events, not whole database
Thank you
for your
attention!
http://commons.wikimedia.org/wiki/File:Preikestolen_Norge.jpg
Slide 9
Moment tensor inversion at
Pyhasalmi ore mine:
accuracy test using explosions
Daniela Kühn (NORSAR)
V. Vavrycuk (Academy of Sciences of the CR)
AIM 2nd annual meeting, 29-30 Sept 2011, Prague, Czech Republic
Pyhäsalmi ore mine, Finland
• microseismic monitoring:
since January 2003
safety of the underground
personnel
optimisation of mining
process
Introduction
Waveform
modelling
Moment
tensor
inversion
• network:
12 1-C geophones
+ 6 3-C geophones (ISS)
3-D geometry
sampling rate: < 3000 Hz
Summary
owned by Inmet Mining Co.
• events:
1500 events /months
(including blasting)
-2 < Mw < 1.5
Inconsistent polarities of P-wave first onset
Moment tensor inversion:
• homogeneous velocity model (as for locations)
• amplitude inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling
Complexity of velocity model
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling: 2D section
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
• E3D: viscoelastic 3-D FD code (Larsen and Schultz, 1995)
• strong interaction with mining cavities: reflection,
scattering, conversion
620 m
• healing of wavefronts
Waveform modelling
synthetic seismograms
- complex waveforms
Introduction
- strong coda
- complex secondary arrivals
Waveform
modelling
- scattering effects stronger
on amplitudes than travel
times,
since
size
of
heterogeneities
(cavities,
access tunnels) same order or
smaller than wavelengths
Moment
tensor
inversion
Summary
arrival times computed by
Eikonal solver still fit
(wavefronts heal quickly
after passing a cavitiy)
observed seismograms
Comparison 1-D/3-D
Imaginary
network!
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Comparison 1-D/3-D
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Influence of proximity to cavity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Real mine
network!
Source depth → ray path
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Ray path → onset polarity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Moment tensor inversion
Amplitude picking
direct wave scattered wave
Introduction
Waveform
modelling
Moment
tensor
inversion
first maximum amplitude = amplitude of the direct wave
direct wave
scattered wave
Summary
energy diffracted
around cavity
?
first maximum amplitude is not always the amplitude of the direct wave
Amplitude vs. waveform inversion
Amplitude inversion:
• homogeneous model of the medium
Introduction
• Green’s functions calculated using ray theory
• inversion of P-wave amplitudes (20-30 amplitudes)
Waveform
modelling
• frequencies: 250-500 Hz
cannot take into account distortion of rays on focal sphere
Moment
tensor
inversion
misinterpretation of amplitudes: which one is the direct wave?
Waveform inversion:
Summary
• 3-D heterogeneous model of the medium
• Green’s functions calculated using FD code
• inversion of full waveforms (15-20 waveforms)
• frequencies (at the moment): 25-100 Hz
inversion is performed in frequency domain
in principle same inversion algorithm as for amplitude inversion, but run
repeatedly for every frequency band (0.5 Hz steps)
Selected explosions: relocation
expl 1
expl 1
expl 3
Introduction
expl 5
Waveform
modelling
expl 3
Moment
tensor
inversion
expl 5
expl 1
Summary
expl 3
expl 5
Explosions (coords in m):
1) x=8306E y=2312N z=-1238
→ 26 m shift
3) x=8218E y=2192N z=-1352
→ 68 m shift
5) x=8214E y=2168N z=-1356
→ 59 m shift
Examples: good fit
Examples: amplitude misfit
Examples: phase misfit
Inversion results
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
explosion 1
explosion 3
explosion 5
Length of time windows
GF duration [10 ms]
Data duration [10 ms]
explosion 1
ISO percentage
GF duration [10 ms]
ISO percentage
GF duration [10 ms]
ISO percentage
Data duration [10 ms]
explosion 3
Data duration [10 ms]
explosion 5
nearly all solutions have high isotropic percentage
best solutions near diagonal
(length of GF time window = length of data time window)
best solutions P –wave + S-wave onset
Explosion 1
Amplitude inversion
Waveform inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
DC = 7%
CLVD = -14%
ISO = 79%
Summary: seismicity in mines
structural model in mines is very complex
Introduction
large and abrupt changes in velocity at
cavities
model varies in time
Waveform
modelling
Moment
tensor
inversion
Summary
earthquake source is complex (single
forces, non-DC components)
small changes in source position lead to
large changes in ray propagation, rays can
be strongly curved
radiated wave field is complex
(reflected, converted, scattered waves, head
waves)
Summary: waveform inversion
In general:
• complex Green’s functions can be calculated by 3-D FD codes (accurate model needed!);
Introduction
• sensitive to time shifts due to mislocation or due to inaccurate velocity model
• frequency band of inverted waves can be easily controlled => stability analysis
Waveform
modelling
In particular:
Moment
tensor
inversion
• good network configuration => focal sphere nicely covered
• inversion algorithm:
• optimal with same window length for Green’s functions and data
Summary
• optimal with simultaneous inversion of P- and S-wave, but excluding S-wave coda
• yields high isotropic percentage, higher than amplitude inversion, almost
independently of window length
promising, but computationally demanding (especially the computation of Green’s
functions with sufficiently small grid point distances)
will be performed for selected events, not whole database
Thank you
for your
attention!
http://commons.wikimedia.org/wiki/File:Preikestolen_Norge.jpg
Slide 10
Moment tensor inversion at
Pyhasalmi ore mine:
accuracy test using explosions
Daniela Kühn (NORSAR)
V. Vavrycuk (Academy of Sciences of the CR)
AIM 2nd annual meeting, 29-30 Sept 2011, Prague, Czech Republic
Pyhäsalmi ore mine, Finland
• microseismic monitoring:
since January 2003
safety of the underground
personnel
optimisation of mining
process
Introduction
Waveform
modelling
Moment
tensor
inversion
• network:
12 1-C geophones
+ 6 3-C geophones (ISS)
3-D geometry
sampling rate: < 3000 Hz
Summary
owned by Inmet Mining Co.
• events:
1500 events /months
(including blasting)
-2 < Mw < 1.5
Inconsistent polarities of P-wave first onset
Moment tensor inversion:
• homogeneous velocity model (as for locations)
• amplitude inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling
Complexity of velocity model
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling: 2D section
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
• E3D: viscoelastic 3-D FD code (Larsen and Schultz, 1995)
• strong interaction with mining cavities: reflection,
scattering, conversion
620 m
• healing of wavefronts
Waveform modelling
synthetic seismograms
- complex waveforms
Introduction
- strong coda
- complex secondary arrivals
Waveform
modelling
- scattering effects stronger
on amplitudes than travel
times,
since
size
of
heterogeneities
(cavities,
access tunnels) same order or
smaller than wavelengths
Moment
tensor
inversion
Summary
arrival times computed by
Eikonal solver still fit
(wavefronts heal quickly
after passing a cavitiy)
observed seismograms
Comparison 1-D/3-D
Imaginary
network!
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Comparison 1-D/3-D
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Influence of proximity to cavity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Real mine
network!
Source depth → ray path
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Ray path → onset polarity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Moment tensor inversion
Amplitude picking
direct wave scattered wave
Introduction
Waveform
modelling
Moment
tensor
inversion
first maximum amplitude = amplitude of the direct wave
direct wave
scattered wave
Summary
energy diffracted
around cavity
?
first maximum amplitude is not always the amplitude of the direct wave
Amplitude vs. waveform inversion
Amplitude inversion:
• homogeneous model of the medium
Introduction
• Green’s functions calculated using ray theory
• inversion of P-wave amplitudes (20-30 amplitudes)
Waveform
modelling
• frequencies: 250-500 Hz
cannot take into account distortion of rays on focal sphere
Moment
tensor
inversion
misinterpretation of amplitudes: which one is the direct wave?
Waveform inversion:
Summary
• 3-D heterogeneous model of the medium
• Green’s functions calculated using FD code
• inversion of full waveforms (15-20 waveforms)
• frequencies (at the moment): 25-100 Hz
inversion is performed in frequency domain
in principle same inversion algorithm as for amplitude inversion, but run
repeatedly for every frequency band (0.5 Hz steps)
Selected explosions: relocation
expl 1
expl 1
expl 3
Introduction
expl 5
Waveform
modelling
expl 3
Moment
tensor
inversion
expl 5
expl 1
Summary
expl 3
expl 5
Explosions (coords in m):
1) x=8306E y=2312N z=-1238
→ 26 m shift
3) x=8218E y=2192N z=-1352
→ 68 m shift
5) x=8214E y=2168N z=-1356
→ 59 m shift
Examples: good fit
Examples: amplitude misfit
Examples: phase misfit
Inversion results
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
explosion 1
explosion 3
explosion 5
Length of time windows
GF duration [10 ms]
Data duration [10 ms]
explosion 1
ISO percentage
GF duration [10 ms]
ISO percentage
GF duration [10 ms]
ISO percentage
Data duration [10 ms]
explosion 3
Data duration [10 ms]
explosion 5
nearly all solutions have high isotropic percentage
best solutions near diagonal
(length of GF time window = length of data time window)
best solutions P –wave + S-wave onset
Explosion 1
Amplitude inversion
Waveform inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
DC = 7%
CLVD = -14%
ISO = 79%
Summary: seismicity in mines
structural model in mines is very complex
Introduction
large and abrupt changes in velocity at
cavities
model varies in time
Waveform
modelling
Moment
tensor
inversion
Summary
earthquake source is complex (single
forces, non-DC components)
small changes in source position lead to
large changes in ray propagation, rays can
be strongly curved
radiated wave field is complex
(reflected, converted, scattered waves, head
waves)
Summary: waveform inversion
In general:
• complex Green’s functions can be calculated by 3-D FD codes (accurate model needed!);
Introduction
• sensitive to time shifts due to mislocation or due to inaccurate velocity model
• frequency band of inverted waves can be easily controlled => stability analysis
Waveform
modelling
In particular:
Moment
tensor
inversion
• good network configuration => focal sphere nicely covered
• inversion algorithm:
• optimal with same window length for Green’s functions and data
Summary
• optimal with simultaneous inversion of P- and S-wave, but excluding S-wave coda
• yields high isotropic percentage, higher than amplitude inversion, almost
independently of window length
promising, but computationally demanding (especially the computation of Green’s
functions with sufficiently small grid point distances)
will be performed for selected events, not whole database
Thank you
for your
attention!
http://commons.wikimedia.org/wiki/File:Preikestolen_Norge.jpg
Slide 11
Moment tensor inversion at
Pyhasalmi ore mine:
accuracy test using explosions
Daniela Kühn (NORSAR)
V. Vavrycuk (Academy of Sciences of the CR)
AIM 2nd annual meeting, 29-30 Sept 2011, Prague, Czech Republic
Pyhäsalmi ore mine, Finland
• microseismic monitoring:
since January 2003
safety of the underground
personnel
optimisation of mining
process
Introduction
Waveform
modelling
Moment
tensor
inversion
• network:
12 1-C geophones
+ 6 3-C geophones (ISS)
3-D geometry
sampling rate: < 3000 Hz
Summary
owned by Inmet Mining Co.
• events:
1500 events /months
(including blasting)
-2 < Mw < 1.5
Inconsistent polarities of P-wave first onset
Moment tensor inversion:
• homogeneous velocity model (as for locations)
• amplitude inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling
Complexity of velocity model
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling: 2D section
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
• E3D: viscoelastic 3-D FD code (Larsen and Schultz, 1995)
• strong interaction with mining cavities: reflection,
scattering, conversion
620 m
• healing of wavefronts
Waveform modelling
synthetic seismograms
- complex waveforms
Introduction
- strong coda
- complex secondary arrivals
Waveform
modelling
- scattering effects stronger
on amplitudes than travel
times,
since
size
of
heterogeneities
(cavities,
access tunnels) same order or
smaller than wavelengths
Moment
tensor
inversion
Summary
arrival times computed by
Eikonal solver still fit
(wavefronts heal quickly
after passing a cavitiy)
observed seismograms
Comparison 1-D/3-D
Imaginary
network!
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Comparison 1-D/3-D
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Influence of proximity to cavity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Real mine
network!
Source depth → ray path
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Ray path → onset polarity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Moment tensor inversion
Amplitude picking
direct wave scattered wave
Introduction
Waveform
modelling
Moment
tensor
inversion
first maximum amplitude = amplitude of the direct wave
direct wave
scattered wave
Summary
energy diffracted
around cavity
?
first maximum amplitude is not always the amplitude of the direct wave
Amplitude vs. waveform inversion
Amplitude inversion:
• homogeneous model of the medium
Introduction
• Green’s functions calculated using ray theory
• inversion of P-wave amplitudes (20-30 amplitudes)
Waveform
modelling
• frequencies: 250-500 Hz
cannot take into account distortion of rays on focal sphere
Moment
tensor
inversion
misinterpretation of amplitudes: which one is the direct wave?
Waveform inversion:
Summary
• 3-D heterogeneous model of the medium
• Green’s functions calculated using FD code
• inversion of full waveforms (15-20 waveforms)
• frequencies (at the moment): 25-100 Hz
inversion is performed in frequency domain
in principle same inversion algorithm as for amplitude inversion, but run
repeatedly for every frequency band (0.5 Hz steps)
Selected explosions: relocation
expl 1
expl 1
expl 3
Introduction
expl 5
Waveform
modelling
expl 3
Moment
tensor
inversion
expl 5
expl 1
Summary
expl 3
expl 5
Explosions (coords in m):
1) x=8306E y=2312N z=-1238
→ 26 m shift
3) x=8218E y=2192N z=-1352
→ 68 m shift
5) x=8214E y=2168N z=-1356
→ 59 m shift
Examples: good fit
Examples: amplitude misfit
Examples: phase misfit
Inversion results
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
explosion 1
explosion 3
explosion 5
Length of time windows
GF duration [10 ms]
Data duration [10 ms]
explosion 1
ISO percentage
GF duration [10 ms]
ISO percentage
GF duration [10 ms]
ISO percentage
Data duration [10 ms]
explosion 3
Data duration [10 ms]
explosion 5
nearly all solutions have high isotropic percentage
best solutions near diagonal
(length of GF time window = length of data time window)
best solutions P –wave + S-wave onset
Explosion 1
Amplitude inversion
Waveform inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
DC = 7%
CLVD = -14%
ISO = 79%
Summary: seismicity in mines
structural model in mines is very complex
Introduction
large and abrupt changes in velocity at
cavities
model varies in time
Waveform
modelling
Moment
tensor
inversion
Summary
earthquake source is complex (single
forces, non-DC components)
small changes in source position lead to
large changes in ray propagation, rays can
be strongly curved
radiated wave field is complex
(reflected, converted, scattered waves, head
waves)
Summary: waveform inversion
In general:
• complex Green’s functions can be calculated by 3-D FD codes (accurate model needed!);
Introduction
• sensitive to time shifts due to mislocation or due to inaccurate velocity model
• frequency band of inverted waves can be easily controlled => stability analysis
Waveform
modelling
In particular:
Moment
tensor
inversion
• good network configuration => focal sphere nicely covered
• inversion algorithm:
• optimal with same window length for Green’s functions and data
Summary
• optimal with simultaneous inversion of P- and S-wave, but excluding S-wave coda
• yields high isotropic percentage, higher than amplitude inversion, almost
independently of window length
promising, but computationally demanding (especially the computation of Green’s
functions with sufficiently small grid point distances)
will be performed for selected events, not whole database
Thank you
for your
attention!
http://commons.wikimedia.org/wiki/File:Preikestolen_Norge.jpg
Slide 12
Moment tensor inversion at
Pyhasalmi ore mine:
accuracy test using explosions
Daniela Kühn (NORSAR)
V. Vavrycuk (Academy of Sciences of the CR)
AIM 2nd annual meeting, 29-30 Sept 2011, Prague, Czech Republic
Pyhäsalmi ore mine, Finland
• microseismic monitoring:
since January 2003
safety of the underground
personnel
optimisation of mining
process
Introduction
Waveform
modelling
Moment
tensor
inversion
• network:
12 1-C geophones
+ 6 3-C geophones (ISS)
3-D geometry
sampling rate: < 3000 Hz
Summary
owned by Inmet Mining Co.
• events:
1500 events /months
(including blasting)
-2 < Mw < 1.5
Inconsistent polarities of P-wave first onset
Moment tensor inversion:
• homogeneous velocity model (as for locations)
• amplitude inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling
Complexity of velocity model
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling: 2D section
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
• E3D: viscoelastic 3-D FD code (Larsen and Schultz, 1995)
• strong interaction with mining cavities: reflection,
scattering, conversion
620 m
• healing of wavefronts
Waveform modelling
synthetic seismograms
- complex waveforms
Introduction
- strong coda
- complex secondary arrivals
Waveform
modelling
- scattering effects stronger
on amplitudes than travel
times,
since
size
of
heterogeneities
(cavities,
access tunnels) same order or
smaller than wavelengths
Moment
tensor
inversion
Summary
arrival times computed by
Eikonal solver still fit
(wavefronts heal quickly
after passing a cavitiy)
observed seismograms
Comparison 1-D/3-D
Imaginary
network!
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Comparison 1-D/3-D
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Influence of proximity to cavity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Real mine
network!
Source depth → ray path
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Ray path → onset polarity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Moment tensor inversion
Amplitude picking
direct wave scattered wave
Introduction
Waveform
modelling
Moment
tensor
inversion
first maximum amplitude = amplitude of the direct wave
direct wave
scattered wave
Summary
energy diffracted
around cavity
?
first maximum amplitude is not always the amplitude of the direct wave
Amplitude vs. waveform inversion
Amplitude inversion:
• homogeneous model of the medium
Introduction
• Green’s functions calculated using ray theory
• inversion of P-wave amplitudes (20-30 amplitudes)
Waveform
modelling
• frequencies: 250-500 Hz
cannot take into account distortion of rays on focal sphere
Moment
tensor
inversion
misinterpretation of amplitudes: which one is the direct wave?
Waveform inversion:
Summary
• 3-D heterogeneous model of the medium
• Green’s functions calculated using FD code
• inversion of full waveforms (15-20 waveforms)
• frequencies (at the moment): 25-100 Hz
inversion is performed in frequency domain
in principle same inversion algorithm as for amplitude inversion, but run
repeatedly for every frequency band (0.5 Hz steps)
Selected explosions: relocation
expl 1
expl 1
expl 3
Introduction
expl 5
Waveform
modelling
expl 3
Moment
tensor
inversion
expl 5
expl 1
Summary
expl 3
expl 5
Explosions (coords in m):
1) x=8306E y=2312N z=-1238
→ 26 m shift
3) x=8218E y=2192N z=-1352
→ 68 m shift
5) x=8214E y=2168N z=-1356
→ 59 m shift
Examples: good fit
Examples: amplitude misfit
Examples: phase misfit
Inversion results
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
explosion 1
explosion 3
explosion 5
Length of time windows
GF duration [10 ms]
Data duration [10 ms]
explosion 1
ISO percentage
GF duration [10 ms]
ISO percentage
GF duration [10 ms]
ISO percentage
Data duration [10 ms]
explosion 3
Data duration [10 ms]
explosion 5
nearly all solutions have high isotropic percentage
best solutions near diagonal
(length of GF time window = length of data time window)
best solutions P –wave + S-wave onset
Explosion 1
Amplitude inversion
Waveform inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
DC = 7%
CLVD = -14%
ISO = 79%
Summary: seismicity in mines
structural model in mines is very complex
Introduction
large and abrupt changes in velocity at
cavities
model varies in time
Waveform
modelling
Moment
tensor
inversion
Summary
earthquake source is complex (single
forces, non-DC components)
small changes in source position lead to
large changes in ray propagation, rays can
be strongly curved
radiated wave field is complex
(reflected, converted, scattered waves, head
waves)
Summary: waveform inversion
In general:
• complex Green’s functions can be calculated by 3-D FD codes (accurate model needed!);
Introduction
• sensitive to time shifts due to mislocation or due to inaccurate velocity model
• frequency band of inverted waves can be easily controlled => stability analysis
Waveform
modelling
In particular:
Moment
tensor
inversion
• good network configuration => focal sphere nicely covered
• inversion algorithm:
• optimal with same window length for Green’s functions and data
Summary
• optimal with simultaneous inversion of P- and S-wave, but excluding S-wave coda
• yields high isotropic percentage, higher than amplitude inversion, almost
independently of window length
promising, but computationally demanding (especially the computation of Green’s
functions with sufficiently small grid point distances)
will be performed for selected events, not whole database
Thank you
for your
attention!
http://commons.wikimedia.org/wiki/File:Preikestolen_Norge.jpg
Slide 13
Moment tensor inversion at
Pyhasalmi ore mine:
accuracy test using explosions
Daniela Kühn (NORSAR)
V. Vavrycuk (Academy of Sciences of the CR)
AIM 2nd annual meeting, 29-30 Sept 2011, Prague, Czech Republic
Pyhäsalmi ore mine, Finland
• microseismic monitoring:
since January 2003
safety of the underground
personnel
optimisation of mining
process
Introduction
Waveform
modelling
Moment
tensor
inversion
• network:
12 1-C geophones
+ 6 3-C geophones (ISS)
3-D geometry
sampling rate: < 3000 Hz
Summary
owned by Inmet Mining Co.
• events:
1500 events /months
(including blasting)
-2 < Mw < 1.5
Inconsistent polarities of P-wave first onset
Moment tensor inversion:
• homogeneous velocity model (as for locations)
• amplitude inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling
Complexity of velocity model
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling: 2D section
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
• E3D: viscoelastic 3-D FD code (Larsen and Schultz, 1995)
• strong interaction with mining cavities: reflection,
scattering, conversion
620 m
• healing of wavefronts
Waveform modelling
synthetic seismograms
- complex waveforms
Introduction
- strong coda
- complex secondary arrivals
Waveform
modelling
- scattering effects stronger
on amplitudes than travel
times,
since
size
of
heterogeneities
(cavities,
access tunnels) same order or
smaller than wavelengths
Moment
tensor
inversion
Summary
arrival times computed by
Eikonal solver still fit
(wavefronts heal quickly
after passing a cavitiy)
observed seismograms
Comparison 1-D/3-D
Imaginary
network!
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Comparison 1-D/3-D
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Influence of proximity to cavity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Real mine
network!
Source depth → ray path
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Ray path → onset polarity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Moment tensor inversion
Amplitude picking
direct wave scattered wave
Introduction
Waveform
modelling
Moment
tensor
inversion
first maximum amplitude = amplitude of the direct wave
direct wave
scattered wave
Summary
energy diffracted
around cavity
?
first maximum amplitude is not always the amplitude of the direct wave
Amplitude vs. waveform inversion
Amplitude inversion:
• homogeneous model of the medium
Introduction
• Green’s functions calculated using ray theory
• inversion of P-wave amplitudes (20-30 amplitudes)
Waveform
modelling
• frequencies: 250-500 Hz
cannot take into account distortion of rays on focal sphere
Moment
tensor
inversion
misinterpretation of amplitudes: which one is the direct wave?
Waveform inversion:
Summary
• 3-D heterogeneous model of the medium
• Green’s functions calculated using FD code
• inversion of full waveforms (15-20 waveforms)
• frequencies (at the moment): 25-100 Hz
inversion is performed in frequency domain
in principle same inversion algorithm as for amplitude inversion, but run
repeatedly for every frequency band (0.5 Hz steps)
Selected explosions: relocation
expl 1
expl 1
expl 3
Introduction
expl 5
Waveform
modelling
expl 3
Moment
tensor
inversion
expl 5
expl 1
Summary
expl 3
expl 5
Explosions (coords in m):
1) x=8306E y=2312N z=-1238
→ 26 m shift
3) x=8218E y=2192N z=-1352
→ 68 m shift
5) x=8214E y=2168N z=-1356
→ 59 m shift
Examples: good fit
Examples: amplitude misfit
Examples: phase misfit
Inversion results
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
explosion 1
explosion 3
explosion 5
Length of time windows
GF duration [10 ms]
Data duration [10 ms]
explosion 1
ISO percentage
GF duration [10 ms]
ISO percentage
GF duration [10 ms]
ISO percentage
Data duration [10 ms]
explosion 3
Data duration [10 ms]
explosion 5
nearly all solutions have high isotropic percentage
best solutions near diagonal
(length of GF time window = length of data time window)
best solutions P –wave + S-wave onset
Explosion 1
Amplitude inversion
Waveform inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
DC = 7%
CLVD = -14%
ISO = 79%
Summary: seismicity in mines
structural model in mines is very complex
Introduction
large and abrupt changes in velocity at
cavities
model varies in time
Waveform
modelling
Moment
tensor
inversion
Summary
earthquake source is complex (single
forces, non-DC components)
small changes in source position lead to
large changes in ray propagation, rays can
be strongly curved
radiated wave field is complex
(reflected, converted, scattered waves, head
waves)
Summary: waveform inversion
In general:
• complex Green’s functions can be calculated by 3-D FD codes (accurate model needed!);
Introduction
• sensitive to time shifts due to mislocation or due to inaccurate velocity model
• frequency band of inverted waves can be easily controlled => stability analysis
Waveform
modelling
In particular:
Moment
tensor
inversion
• good network configuration => focal sphere nicely covered
• inversion algorithm:
• optimal with same window length for Green’s functions and data
Summary
• optimal with simultaneous inversion of P- and S-wave, but excluding S-wave coda
• yields high isotropic percentage, higher than amplitude inversion, almost
independently of window length
promising, but computationally demanding (especially the computation of Green’s
functions with sufficiently small grid point distances)
will be performed for selected events, not whole database
Thank you
for your
attention!
http://commons.wikimedia.org/wiki/File:Preikestolen_Norge.jpg
Slide 14
Moment tensor inversion at
Pyhasalmi ore mine:
accuracy test using explosions
Daniela Kühn (NORSAR)
V. Vavrycuk (Academy of Sciences of the CR)
AIM 2nd annual meeting, 29-30 Sept 2011, Prague, Czech Republic
Pyhäsalmi ore mine, Finland
• microseismic monitoring:
since January 2003
safety of the underground
personnel
optimisation of mining
process
Introduction
Waveform
modelling
Moment
tensor
inversion
• network:
12 1-C geophones
+ 6 3-C geophones (ISS)
3-D geometry
sampling rate: < 3000 Hz
Summary
owned by Inmet Mining Co.
• events:
1500 events /months
(including blasting)
-2 < Mw < 1.5
Inconsistent polarities of P-wave first onset
Moment tensor inversion:
• homogeneous velocity model (as for locations)
• amplitude inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling
Complexity of velocity model
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling: 2D section
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
• E3D: viscoelastic 3-D FD code (Larsen and Schultz, 1995)
• strong interaction with mining cavities: reflection,
scattering, conversion
620 m
• healing of wavefronts
Waveform modelling
synthetic seismograms
- complex waveforms
Introduction
- strong coda
- complex secondary arrivals
Waveform
modelling
- scattering effects stronger
on amplitudes than travel
times,
since
size
of
heterogeneities
(cavities,
access tunnels) same order or
smaller than wavelengths
Moment
tensor
inversion
Summary
arrival times computed by
Eikonal solver still fit
(wavefronts heal quickly
after passing a cavitiy)
observed seismograms
Comparison 1-D/3-D
Imaginary
network!
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Comparison 1-D/3-D
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Influence of proximity to cavity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Real mine
network!
Source depth → ray path
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Ray path → onset polarity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Moment tensor inversion
Amplitude picking
direct wave scattered wave
Introduction
Waveform
modelling
Moment
tensor
inversion
first maximum amplitude = amplitude of the direct wave
direct wave
scattered wave
Summary
energy diffracted
around cavity
?
first maximum amplitude is not always the amplitude of the direct wave
Amplitude vs. waveform inversion
Amplitude inversion:
• homogeneous model of the medium
Introduction
• Green’s functions calculated using ray theory
• inversion of P-wave amplitudes (20-30 amplitudes)
Waveform
modelling
• frequencies: 250-500 Hz
cannot take into account distortion of rays on focal sphere
Moment
tensor
inversion
misinterpretation of amplitudes: which one is the direct wave?
Waveform inversion:
Summary
• 3-D heterogeneous model of the medium
• Green’s functions calculated using FD code
• inversion of full waveforms (15-20 waveforms)
• frequencies (at the moment): 25-100 Hz
inversion is performed in frequency domain
in principle same inversion algorithm as for amplitude inversion, but run
repeatedly for every frequency band (0.5 Hz steps)
Selected explosions: relocation
expl 1
expl 1
expl 3
Introduction
expl 5
Waveform
modelling
expl 3
Moment
tensor
inversion
expl 5
expl 1
Summary
expl 3
expl 5
Explosions (coords in m):
1) x=8306E y=2312N z=-1238
→ 26 m shift
3) x=8218E y=2192N z=-1352
→ 68 m shift
5) x=8214E y=2168N z=-1356
→ 59 m shift
Examples: good fit
Examples: amplitude misfit
Examples: phase misfit
Inversion results
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
explosion 1
explosion 3
explosion 5
Length of time windows
GF duration [10 ms]
Data duration [10 ms]
explosion 1
ISO percentage
GF duration [10 ms]
ISO percentage
GF duration [10 ms]
ISO percentage
Data duration [10 ms]
explosion 3
Data duration [10 ms]
explosion 5
nearly all solutions have high isotropic percentage
best solutions near diagonal
(length of GF time window = length of data time window)
best solutions P –wave + S-wave onset
Explosion 1
Amplitude inversion
Waveform inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
DC = 7%
CLVD = -14%
ISO = 79%
Summary: seismicity in mines
structural model in mines is very complex
Introduction
large and abrupt changes in velocity at
cavities
model varies in time
Waveform
modelling
Moment
tensor
inversion
Summary
earthquake source is complex (single
forces, non-DC components)
small changes in source position lead to
large changes in ray propagation, rays can
be strongly curved
radiated wave field is complex
(reflected, converted, scattered waves, head
waves)
Summary: waveform inversion
In general:
• complex Green’s functions can be calculated by 3-D FD codes (accurate model needed!);
Introduction
• sensitive to time shifts due to mislocation or due to inaccurate velocity model
• frequency band of inverted waves can be easily controlled => stability analysis
Waveform
modelling
In particular:
Moment
tensor
inversion
• good network configuration => focal sphere nicely covered
• inversion algorithm:
• optimal with same window length for Green’s functions and data
Summary
• optimal with simultaneous inversion of P- and S-wave, but excluding S-wave coda
• yields high isotropic percentage, higher than amplitude inversion, almost
independently of window length
promising, but computationally demanding (especially the computation of Green’s
functions with sufficiently small grid point distances)
will be performed for selected events, not whole database
Thank you
for your
attention!
http://commons.wikimedia.org/wiki/File:Preikestolen_Norge.jpg
Slide 15
Moment tensor inversion at
Pyhasalmi ore mine:
accuracy test using explosions
Daniela Kühn (NORSAR)
V. Vavrycuk (Academy of Sciences of the CR)
AIM 2nd annual meeting, 29-30 Sept 2011, Prague, Czech Republic
Pyhäsalmi ore mine, Finland
• microseismic monitoring:
since January 2003
safety of the underground
personnel
optimisation of mining
process
Introduction
Waveform
modelling
Moment
tensor
inversion
• network:
12 1-C geophones
+ 6 3-C geophones (ISS)
3-D geometry
sampling rate: < 3000 Hz
Summary
owned by Inmet Mining Co.
• events:
1500 events /months
(including blasting)
-2 < Mw < 1.5
Inconsistent polarities of P-wave first onset
Moment tensor inversion:
• homogeneous velocity model (as for locations)
• amplitude inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling
Complexity of velocity model
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling: 2D section
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
• E3D: viscoelastic 3-D FD code (Larsen and Schultz, 1995)
• strong interaction with mining cavities: reflection,
scattering, conversion
620 m
• healing of wavefronts
Waveform modelling
synthetic seismograms
- complex waveforms
Introduction
- strong coda
- complex secondary arrivals
Waveform
modelling
- scattering effects stronger
on amplitudes than travel
times,
since
size
of
heterogeneities
(cavities,
access tunnels) same order or
smaller than wavelengths
Moment
tensor
inversion
Summary
arrival times computed by
Eikonal solver still fit
(wavefronts heal quickly
after passing a cavitiy)
observed seismograms
Comparison 1-D/3-D
Imaginary
network!
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Comparison 1-D/3-D
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Influence of proximity to cavity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Real mine
network!
Source depth → ray path
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Ray path → onset polarity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Moment tensor inversion
Amplitude picking
direct wave scattered wave
Introduction
Waveform
modelling
Moment
tensor
inversion
first maximum amplitude = amplitude of the direct wave
direct wave
scattered wave
Summary
energy diffracted
around cavity
?
first maximum amplitude is not always the amplitude of the direct wave
Amplitude vs. waveform inversion
Amplitude inversion:
• homogeneous model of the medium
Introduction
• Green’s functions calculated using ray theory
• inversion of P-wave amplitudes (20-30 amplitudes)
Waveform
modelling
• frequencies: 250-500 Hz
cannot take into account distortion of rays on focal sphere
Moment
tensor
inversion
misinterpretation of amplitudes: which one is the direct wave?
Waveform inversion:
Summary
• 3-D heterogeneous model of the medium
• Green’s functions calculated using FD code
• inversion of full waveforms (15-20 waveforms)
• frequencies (at the moment): 25-100 Hz
inversion is performed in frequency domain
in principle same inversion algorithm as for amplitude inversion, but run
repeatedly for every frequency band (0.5 Hz steps)
Selected explosions: relocation
expl 1
expl 1
expl 3
Introduction
expl 5
Waveform
modelling
expl 3
Moment
tensor
inversion
expl 5
expl 1
Summary
expl 3
expl 5
Explosions (coords in m):
1) x=8306E y=2312N z=-1238
→ 26 m shift
3) x=8218E y=2192N z=-1352
→ 68 m shift
5) x=8214E y=2168N z=-1356
→ 59 m shift
Examples: good fit
Examples: amplitude misfit
Examples: phase misfit
Inversion results
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
explosion 1
explosion 3
explosion 5
Length of time windows
GF duration [10 ms]
Data duration [10 ms]
explosion 1
ISO percentage
GF duration [10 ms]
ISO percentage
GF duration [10 ms]
ISO percentage
Data duration [10 ms]
explosion 3
Data duration [10 ms]
explosion 5
nearly all solutions have high isotropic percentage
best solutions near diagonal
(length of GF time window = length of data time window)
best solutions P –wave + S-wave onset
Explosion 1
Amplitude inversion
Waveform inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
DC = 7%
CLVD = -14%
ISO = 79%
Summary: seismicity in mines
structural model in mines is very complex
Introduction
large and abrupt changes in velocity at
cavities
model varies in time
Waveform
modelling
Moment
tensor
inversion
Summary
earthquake source is complex (single
forces, non-DC components)
small changes in source position lead to
large changes in ray propagation, rays can
be strongly curved
radiated wave field is complex
(reflected, converted, scattered waves, head
waves)
Summary: waveform inversion
In general:
• complex Green’s functions can be calculated by 3-D FD codes (accurate model needed!);
Introduction
• sensitive to time shifts due to mislocation or due to inaccurate velocity model
• frequency band of inverted waves can be easily controlled => stability analysis
Waveform
modelling
In particular:
Moment
tensor
inversion
• good network configuration => focal sphere nicely covered
• inversion algorithm:
• optimal with same window length for Green’s functions and data
Summary
• optimal with simultaneous inversion of P- and S-wave, but excluding S-wave coda
• yields high isotropic percentage, higher than amplitude inversion, almost
independently of window length
promising, but computationally demanding (especially the computation of Green’s
functions with sufficiently small grid point distances)
will be performed for selected events, not whole database
Thank you
for your
attention!
http://commons.wikimedia.org/wiki/File:Preikestolen_Norge.jpg
Slide 16
Moment tensor inversion at
Pyhasalmi ore mine:
accuracy test using explosions
Daniela Kühn (NORSAR)
V. Vavrycuk (Academy of Sciences of the CR)
AIM 2nd annual meeting, 29-30 Sept 2011, Prague, Czech Republic
Pyhäsalmi ore mine, Finland
• microseismic monitoring:
since January 2003
safety of the underground
personnel
optimisation of mining
process
Introduction
Waveform
modelling
Moment
tensor
inversion
• network:
12 1-C geophones
+ 6 3-C geophones (ISS)
3-D geometry
sampling rate: < 3000 Hz
Summary
owned by Inmet Mining Co.
• events:
1500 events /months
(including blasting)
-2 < Mw < 1.5
Inconsistent polarities of P-wave first onset
Moment tensor inversion:
• homogeneous velocity model (as for locations)
• amplitude inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling
Complexity of velocity model
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling: 2D section
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
• E3D: viscoelastic 3-D FD code (Larsen and Schultz, 1995)
• strong interaction with mining cavities: reflection,
scattering, conversion
620 m
• healing of wavefronts
Waveform modelling
synthetic seismograms
- complex waveforms
Introduction
- strong coda
- complex secondary arrivals
Waveform
modelling
- scattering effects stronger
on amplitudes than travel
times,
since
size
of
heterogeneities
(cavities,
access tunnels) same order or
smaller than wavelengths
Moment
tensor
inversion
Summary
arrival times computed by
Eikonal solver still fit
(wavefronts heal quickly
after passing a cavitiy)
observed seismograms
Comparison 1-D/3-D
Imaginary
network!
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Comparison 1-D/3-D
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Influence of proximity to cavity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Real mine
network!
Source depth → ray path
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Ray path → onset polarity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Moment tensor inversion
Amplitude picking
direct wave scattered wave
Introduction
Waveform
modelling
Moment
tensor
inversion
first maximum amplitude = amplitude of the direct wave
direct wave
scattered wave
Summary
energy diffracted
around cavity
?
first maximum amplitude is not always the amplitude of the direct wave
Amplitude vs. waveform inversion
Amplitude inversion:
• homogeneous model of the medium
Introduction
• Green’s functions calculated using ray theory
• inversion of P-wave amplitudes (20-30 amplitudes)
Waveform
modelling
• frequencies: 250-500 Hz
cannot take into account distortion of rays on focal sphere
Moment
tensor
inversion
misinterpretation of amplitudes: which one is the direct wave?
Waveform inversion:
Summary
• 3-D heterogeneous model of the medium
• Green’s functions calculated using FD code
• inversion of full waveforms (15-20 waveforms)
• frequencies (at the moment): 25-100 Hz
inversion is performed in frequency domain
in principle same inversion algorithm as for amplitude inversion, but run
repeatedly for every frequency band (0.5 Hz steps)
Selected explosions: relocation
expl 1
expl 1
expl 3
Introduction
expl 5
Waveform
modelling
expl 3
Moment
tensor
inversion
expl 5
expl 1
Summary
expl 3
expl 5
Explosions (coords in m):
1) x=8306E y=2312N z=-1238
→ 26 m shift
3) x=8218E y=2192N z=-1352
→ 68 m shift
5) x=8214E y=2168N z=-1356
→ 59 m shift
Examples: good fit
Examples: amplitude misfit
Examples: phase misfit
Inversion results
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
explosion 1
explosion 3
explosion 5
Length of time windows
GF duration [10 ms]
Data duration [10 ms]
explosion 1
ISO percentage
GF duration [10 ms]
ISO percentage
GF duration [10 ms]
ISO percentage
Data duration [10 ms]
explosion 3
Data duration [10 ms]
explosion 5
nearly all solutions have high isotropic percentage
best solutions near diagonal
(length of GF time window = length of data time window)
best solutions P –wave + S-wave onset
Explosion 1
Amplitude inversion
Waveform inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
DC = 7%
CLVD = -14%
ISO = 79%
Summary: seismicity in mines
structural model in mines is very complex
Introduction
large and abrupt changes in velocity at
cavities
model varies in time
Waveform
modelling
Moment
tensor
inversion
Summary
earthquake source is complex (single
forces, non-DC components)
small changes in source position lead to
large changes in ray propagation, rays can
be strongly curved
radiated wave field is complex
(reflected, converted, scattered waves, head
waves)
Summary: waveform inversion
In general:
• complex Green’s functions can be calculated by 3-D FD codes (accurate model needed!);
Introduction
• sensitive to time shifts due to mislocation or due to inaccurate velocity model
• frequency band of inverted waves can be easily controlled => stability analysis
Waveform
modelling
In particular:
Moment
tensor
inversion
• good network configuration => focal sphere nicely covered
• inversion algorithm:
• optimal with same window length for Green’s functions and data
Summary
• optimal with simultaneous inversion of P- and S-wave, but excluding S-wave coda
• yields high isotropic percentage, higher than amplitude inversion, almost
independently of window length
promising, but computationally demanding (especially the computation of Green’s
functions with sufficiently small grid point distances)
will be performed for selected events, not whole database
Thank you
for your
attention!
http://commons.wikimedia.org/wiki/File:Preikestolen_Norge.jpg
Slide 17
Moment tensor inversion at
Pyhasalmi ore mine:
accuracy test using explosions
Daniela Kühn (NORSAR)
V. Vavrycuk (Academy of Sciences of the CR)
AIM 2nd annual meeting, 29-30 Sept 2011, Prague, Czech Republic
Pyhäsalmi ore mine, Finland
• microseismic monitoring:
since January 2003
safety of the underground
personnel
optimisation of mining
process
Introduction
Waveform
modelling
Moment
tensor
inversion
• network:
12 1-C geophones
+ 6 3-C geophones (ISS)
3-D geometry
sampling rate: < 3000 Hz
Summary
owned by Inmet Mining Co.
• events:
1500 events /months
(including blasting)
-2 < Mw < 1.5
Inconsistent polarities of P-wave first onset
Moment tensor inversion:
• homogeneous velocity model (as for locations)
• amplitude inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling
Complexity of velocity model
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling: 2D section
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
• E3D: viscoelastic 3-D FD code (Larsen and Schultz, 1995)
• strong interaction with mining cavities: reflection,
scattering, conversion
620 m
• healing of wavefronts
Waveform modelling
synthetic seismograms
- complex waveforms
Introduction
- strong coda
- complex secondary arrivals
Waveform
modelling
- scattering effects stronger
on amplitudes than travel
times,
since
size
of
heterogeneities
(cavities,
access tunnels) same order or
smaller than wavelengths
Moment
tensor
inversion
Summary
arrival times computed by
Eikonal solver still fit
(wavefronts heal quickly
after passing a cavitiy)
observed seismograms
Comparison 1-D/3-D
Imaginary
network!
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Comparison 1-D/3-D
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Influence of proximity to cavity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Real mine
network!
Source depth → ray path
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Ray path → onset polarity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Moment tensor inversion
Amplitude picking
direct wave scattered wave
Introduction
Waveform
modelling
Moment
tensor
inversion
first maximum amplitude = amplitude of the direct wave
direct wave
scattered wave
Summary
energy diffracted
around cavity
?
first maximum amplitude is not always the amplitude of the direct wave
Amplitude vs. waveform inversion
Amplitude inversion:
• homogeneous model of the medium
Introduction
• Green’s functions calculated using ray theory
• inversion of P-wave amplitudes (20-30 amplitudes)
Waveform
modelling
• frequencies: 250-500 Hz
cannot take into account distortion of rays on focal sphere
Moment
tensor
inversion
misinterpretation of amplitudes: which one is the direct wave?
Waveform inversion:
Summary
• 3-D heterogeneous model of the medium
• Green’s functions calculated using FD code
• inversion of full waveforms (15-20 waveforms)
• frequencies (at the moment): 25-100 Hz
inversion is performed in frequency domain
in principle same inversion algorithm as for amplitude inversion, but run
repeatedly for every frequency band (0.5 Hz steps)
Selected explosions: relocation
expl 1
expl 1
expl 3
Introduction
expl 5
Waveform
modelling
expl 3
Moment
tensor
inversion
expl 5
expl 1
Summary
expl 3
expl 5
Explosions (coords in m):
1) x=8306E y=2312N z=-1238
→ 26 m shift
3) x=8218E y=2192N z=-1352
→ 68 m shift
5) x=8214E y=2168N z=-1356
→ 59 m shift
Examples: good fit
Examples: amplitude misfit
Examples: phase misfit
Inversion results
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
explosion 1
explosion 3
explosion 5
Length of time windows
GF duration [10 ms]
Data duration [10 ms]
explosion 1
ISO percentage
GF duration [10 ms]
ISO percentage
GF duration [10 ms]
ISO percentage
Data duration [10 ms]
explosion 3
Data duration [10 ms]
explosion 5
nearly all solutions have high isotropic percentage
best solutions near diagonal
(length of GF time window = length of data time window)
best solutions P –wave + S-wave onset
Explosion 1
Amplitude inversion
Waveform inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
DC = 7%
CLVD = -14%
ISO = 79%
Summary: seismicity in mines
structural model in mines is very complex
Introduction
large and abrupt changes in velocity at
cavities
model varies in time
Waveform
modelling
Moment
tensor
inversion
Summary
earthquake source is complex (single
forces, non-DC components)
small changes in source position lead to
large changes in ray propagation, rays can
be strongly curved
radiated wave field is complex
(reflected, converted, scattered waves, head
waves)
Summary: waveform inversion
In general:
• complex Green’s functions can be calculated by 3-D FD codes (accurate model needed!);
Introduction
• sensitive to time shifts due to mislocation or due to inaccurate velocity model
• frequency band of inverted waves can be easily controlled => stability analysis
Waveform
modelling
In particular:
Moment
tensor
inversion
• good network configuration => focal sphere nicely covered
• inversion algorithm:
• optimal with same window length for Green’s functions and data
Summary
• optimal with simultaneous inversion of P- and S-wave, but excluding S-wave coda
• yields high isotropic percentage, higher than amplitude inversion, almost
independently of window length
promising, but computationally demanding (especially the computation of Green’s
functions with sufficiently small grid point distances)
will be performed for selected events, not whole database
Thank you
for your
attention!
http://commons.wikimedia.org/wiki/File:Preikestolen_Norge.jpg
Slide 18
Moment tensor inversion at
Pyhasalmi ore mine:
accuracy test using explosions
Daniela Kühn (NORSAR)
V. Vavrycuk (Academy of Sciences of the CR)
AIM 2nd annual meeting, 29-30 Sept 2011, Prague, Czech Republic
Pyhäsalmi ore mine, Finland
• microseismic monitoring:
since January 2003
safety of the underground
personnel
optimisation of mining
process
Introduction
Waveform
modelling
Moment
tensor
inversion
• network:
12 1-C geophones
+ 6 3-C geophones (ISS)
3-D geometry
sampling rate: < 3000 Hz
Summary
owned by Inmet Mining Co.
• events:
1500 events /months
(including blasting)
-2 < Mw < 1.5
Inconsistent polarities of P-wave first onset
Moment tensor inversion:
• homogeneous velocity model (as for locations)
• amplitude inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling
Complexity of velocity model
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling: 2D section
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
• E3D: viscoelastic 3-D FD code (Larsen and Schultz, 1995)
• strong interaction with mining cavities: reflection,
scattering, conversion
620 m
• healing of wavefronts
Waveform modelling
synthetic seismograms
- complex waveforms
Introduction
- strong coda
- complex secondary arrivals
Waveform
modelling
- scattering effects stronger
on amplitudes than travel
times,
since
size
of
heterogeneities
(cavities,
access tunnels) same order or
smaller than wavelengths
Moment
tensor
inversion
Summary
arrival times computed by
Eikonal solver still fit
(wavefronts heal quickly
after passing a cavitiy)
observed seismograms
Comparison 1-D/3-D
Imaginary
network!
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Comparison 1-D/3-D
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Influence of proximity to cavity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Real mine
network!
Source depth → ray path
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Ray path → onset polarity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Moment tensor inversion
Amplitude picking
direct wave scattered wave
Introduction
Waveform
modelling
Moment
tensor
inversion
first maximum amplitude = amplitude of the direct wave
direct wave
scattered wave
Summary
energy diffracted
around cavity
?
first maximum amplitude is not always the amplitude of the direct wave
Amplitude vs. waveform inversion
Amplitude inversion:
• homogeneous model of the medium
Introduction
• Green’s functions calculated using ray theory
• inversion of P-wave amplitudes (20-30 amplitudes)
Waveform
modelling
• frequencies: 250-500 Hz
cannot take into account distortion of rays on focal sphere
Moment
tensor
inversion
misinterpretation of amplitudes: which one is the direct wave?
Waveform inversion:
Summary
• 3-D heterogeneous model of the medium
• Green’s functions calculated using FD code
• inversion of full waveforms (15-20 waveforms)
• frequencies (at the moment): 25-100 Hz
inversion is performed in frequency domain
in principle same inversion algorithm as for amplitude inversion, but run
repeatedly for every frequency band (0.5 Hz steps)
Selected explosions: relocation
expl 1
expl 1
expl 3
Introduction
expl 5
Waveform
modelling
expl 3
Moment
tensor
inversion
expl 5
expl 1
Summary
expl 3
expl 5
Explosions (coords in m):
1) x=8306E y=2312N z=-1238
→ 26 m shift
3) x=8218E y=2192N z=-1352
→ 68 m shift
5) x=8214E y=2168N z=-1356
→ 59 m shift
Examples: good fit
Examples: amplitude misfit
Examples: phase misfit
Inversion results
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
explosion 1
explosion 3
explosion 5
Length of time windows
GF duration [10 ms]
Data duration [10 ms]
explosion 1
ISO percentage
GF duration [10 ms]
ISO percentage
GF duration [10 ms]
ISO percentage
Data duration [10 ms]
explosion 3
Data duration [10 ms]
explosion 5
nearly all solutions have high isotropic percentage
best solutions near diagonal
(length of GF time window = length of data time window)
best solutions P –wave + S-wave onset
Explosion 1
Amplitude inversion
Waveform inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
DC = 7%
CLVD = -14%
ISO = 79%
Summary: seismicity in mines
structural model in mines is very complex
Introduction
large and abrupt changes in velocity at
cavities
model varies in time
Waveform
modelling
Moment
tensor
inversion
Summary
earthquake source is complex (single
forces, non-DC components)
small changes in source position lead to
large changes in ray propagation, rays can
be strongly curved
radiated wave field is complex
(reflected, converted, scattered waves, head
waves)
Summary: waveform inversion
In general:
• complex Green’s functions can be calculated by 3-D FD codes (accurate model needed!);
Introduction
• sensitive to time shifts due to mislocation or due to inaccurate velocity model
• frequency band of inverted waves can be easily controlled => stability analysis
Waveform
modelling
In particular:
Moment
tensor
inversion
• good network configuration => focal sphere nicely covered
• inversion algorithm:
• optimal with same window length for Green’s functions and data
Summary
• optimal with simultaneous inversion of P- and S-wave, but excluding S-wave coda
• yields high isotropic percentage, higher than amplitude inversion, almost
independently of window length
promising, but computationally demanding (especially the computation of Green’s
functions with sufficiently small grid point distances)
will be performed for selected events, not whole database
Thank you
for your
attention!
http://commons.wikimedia.org/wiki/File:Preikestolen_Norge.jpg
Slide 19
Moment tensor inversion at
Pyhasalmi ore mine:
accuracy test using explosions
Daniela Kühn (NORSAR)
V. Vavrycuk (Academy of Sciences of the CR)
AIM 2nd annual meeting, 29-30 Sept 2011, Prague, Czech Republic
Pyhäsalmi ore mine, Finland
• microseismic monitoring:
since January 2003
safety of the underground
personnel
optimisation of mining
process
Introduction
Waveform
modelling
Moment
tensor
inversion
• network:
12 1-C geophones
+ 6 3-C geophones (ISS)
3-D geometry
sampling rate: < 3000 Hz
Summary
owned by Inmet Mining Co.
• events:
1500 events /months
(including blasting)
-2 < Mw < 1.5
Inconsistent polarities of P-wave first onset
Moment tensor inversion:
• homogeneous velocity model (as for locations)
• amplitude inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling
Complexity of velocity model
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling: 2D section
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
• E3D: viscoelastic 3-D FD code (Larsen and Schultz, 1995)
• strong interaction with mining cavities: reflection,
scattering, conversion
620 m
• healing of wavefronts
Waveform modelling
synthetic seismograms
- complex waveforms
Introduction
- strong coda
- complex secondary arrivals
Waveform
modelling
- scattering effects stronger
on amplitudes than travel
times,
since
size
of
heterogeneities
(cavities,
access tunnels) same order or
smaller than wavelengths
Moment
tensor
inversion
Summary
arrival times computed by
Eikonal solver still fit
(wavefronts heal quickly
after passing a cavitiy)
observed seismograms
Comparison 1-D/3-D
Imaginary
network!
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Comparison 1-D/3-D
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Influence of proximity to cavity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Real mine
network!
Source depth → ray path
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Ray path → onset polarity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Moment tensor inversion
Amplitude picking
direct wave scattered wave
Introduction
Waveform
modelling
Moment
tensor
inversion
first maximum amplitude = amplitude of the direct wave
direct wave
scattered wave
Summary
energy diffracted
around cavity
?
first maximum amplitude is not always the amplitude of the direct wave
Amplitude vs. waveform inversion
Amplitude inversion:
• homogeneous model of the medium
Introduction
• Green’s functions calculated using ray theory
• inversion of P-wave amplitudes (20-30 amplitudes)
Waveform
modelling
• frequencies: 250-500 Hz
cannot take into account distortion of rays on focal sphere
Moment
tensor
inversion
misinterpretation of amplitudes: which one is the direct wave?
Waveform inversion:
Summary
• 3-D heterogeneous model of the medium
• Green’s functions calculated using FD code
• inversion of full waveforms (15-20 waveforms)
• frequencies (at the moment): 25-100 Hz
inversion is performed in frequency domain
in principle same inversion algorithm as for amplitude inversion, but run
repeatedly for every frequency band (0.5 Hz steps)
Selected explosions: relocation
expl 1
expl 1
expl 3
Introduction
expl 5
Waveform
modelling
expl 3
Moment
tensor
inversion
expl 5
expl 1
Summary
expl 3
expl 5
Explosions (coords in m):
1) x=8306E y=2312N z=-1238
→ 26 m shift
3) x=8218E y=2192N z=-1352
→ 68 m shift
5) x=8214E y=2168N z=-1356
→ 59 m shift
Examples: good fit
Examples: amplitude misfit
Examples: phase misfit
Inversion results
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
explosion 1
explosion 3
explosion 5
Length of time windows
GF duration [10 ms]
Data duration [10 ms]
explosion 1
ISO percentage
GF duration [10 ms]
ISO percentage
GF duration [10 ms]
ISO percentage
Data duration [10 ms]
explosion 3
Data duration [10 ms]
explosion 5
nearly all solutions have high isotropic percentage
best solutions near diagonal
(length of GF time window = length of data time window)
best solutions P –wave + S-wave onset
Explosion 1
Amplitude inversion
Waveform inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
DC = 7%
CLVD = -14%
ISO = 79%
Summary: seismicity in mines
structural model in mines is very complex
Introduction
large and abrupt changes in velocity at
cavities
model varies in time
Waveform
modelling
Moment
tensor
inversion
Summary
earthquake source is complex (single
forces, non-DC components)
small changes in source position lead to
large changes in ray propagation, rays can
be strongly curved
radiated wave field is complex
(reflected, converted, scattered waves, head
waves)
Summary: waveform inversion
In general:
• complex Green’s functions can be calculated by 3-D FD codes (accurate model needed!);
Introduction
• sensitive to time shifts due to mislocation or due to inaccurate velocity model
• frequency band of inverted waves can be easily controlled => stability analysis
Waveform
modelling
In particular:
Moment
tensor
inversion
• good network configuration => focal sphere nicely covered
• inversion algorithm:
• optimal with same window length for Green’s functions and data
Summary
• optimal with simultaneous inversion of P- and S-wave, but excluding S-wave coda
• yields high isotropic percentage, higher than amplitude inversion, almost
independently of window length
promising, but computationally demanding (especially the computation of Green’s
functions with sufficiently small grid point distances)
will be performed for selected events, not whole database
Thank you
for your
attention!
http://commons.wikimedia.org/wiki/File:Preikestolen_Norge.jpg
Slide 20
Moment tensor inversion at
Pyhasalmi ore mine:
accuracy test using explosions
Daniela Kühn (NORSAR)
V. Vavrycuk (Academy of Sciences of the CR)
AIM 2nd annual meeting, 29-30 Sept 2011, Prague, Czech Republic
Pyhäsalmi ore mine, Finland
• microseismic monitoring:
since January 2003
safety of the underground
personnel
optimisation of mining
process
Introduction
Waveform
modelling
Moment
tensor
inversion
• network:
12 1-C geophones
+ 6 3-C geophones (ISS)
3-D geometry
sampling rate: < 3000 Hz
Summary
owned by Inmet Mining Co.
• events:
1500 events /months
(including blasting)
-2 < Mw < 1.5
Inconsistent polarities of P-wave first onset
Moment tensor inversion:
• homogeneous velocity model (as for locations)
• amplitude inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling
Complexity of velocity model
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling: 2D section
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
• E3D: viscoelastic 3-D FD code (Larsen and Schultz, 1995)
• strong interaction with mining cavities: reflection,
scattering, conversion
620 m
• healing of wavefronts
Waveform modelling
synthetic seismograms
- complex waveforms
Introduction
- strong coda
- complex secondary arrivals
Waveform
modelling
- scattering effects stronger
on amplitudes than travel
times,
since
size
of
heterogeneities
(cavities,
access tunnels) same order or
smaller than wavelengths
Moment
tensor
inversion
Summary
arrival times computed by
Eikonal solver still fit
(wavefronts heal quickly
after passing a cavitiy)
observed seismograms
Comparison 1-D/3-D
Imaginary
network!
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Comparison 1-D/3-D
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Influence of proximity to cavity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Real mine
network!
Source depth → ray path
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Ray path → onset polarity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Moment tensor inversion
Amplitude picking
direct wave scattered wave
Introduction
Waveform
modelling
Moment
tensor
inversion
first maximum amplitude = amplitude of the direct wave
direct wave
scattered wave
Summary
energy diffracted
around cavity
?
first maximum amplitude is not always the amplitude of the direct wave
Amplitude vs. waveform inversion
Amplitude inversion:
• homogeneous model of the medium
Introduction
• Green’s functions calculated using ray theory
• inversion of P-wave amplitudes (20-30 amplitudes)
Waveform
modelling
• frequencies: 250-500 Hz
cannot take into account distortion of rays on focal sphere
Moment
tensor
inversion
misinterpretation of amplitudes: which one is the direct wave?
Waveform inversion:
Summary
• 3-D heterogeneous model of the medium
• Green’s functions calculated using FD code
• inversion of full waveforms (15-20 waveforms)
• frequencies (at the moment): 25-100 Hz
inversion is performed in frequency domain
in principle same inversion algorithm as for amplitude inversion, but run
repeatedly for every frequency band (0.5 Hz steps)
Selected explosions: relocation
expl 1
expl 1
expl 3
Introduction
expl 5
Waveform
modelling
expl 3
Moment
tensor
inversion
expl 5
expl 1
Summary
expl 3
expl 5
Explosions (coords in m):
1) x=8306E y=2312N z=-1238
→ 26 m shift
3) x=8218E y=2192N z=-1352
→ 68 m shift
5) x=8214E y=2168N z=-1356
→ 59 m shift
Examples: good fit
Examples: amplitude misfit
Examples: phase misfit
Inversion results
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
explosion 1
explosion 3
explosion 5
Length of time windows
GF duration [10 ms]
Data duration [10 ms]
explosion 1
ISO percentage
GF duration [10 ms]
ISO percentage
GF duration [10 ms]
ISO percentage
Data duration [10 ms]
explosion 3
Data duration [10 ms]
explosion 5
nearly all solutions have high isotropic percentage
best solutions near diagonal
(length of GF time window = length of data time window)
best solutions P –wave + S-wave onset
Explosion 1
Amplitude inversion
Waveform inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
DC = 7%
CLVD = -14%
ISO = 79%
Summary: seismicity in mines
structural model in mines is very complex
Introduction
large and abrupt changes in velocity at
cavities
model varies in time
Waveform
modelling
Moment
tensor
inversion
Summary
earthquake source is complex (single
forces, non-DC components)
small changes in source position lead to
large changes in ray propagation, rays can
be strongly curved
radiated wave field is complex
(reflected, converted, scattered waves, head
waves)
Summary: waveform inversion
In general:
• complex Green’s functions can be calculated by 3-D FD codes (accurate model needed!);
Introduction
• sensitive to time shifts due to mislocation or due to inaccurate velocity model
• frequency band of inverted waves can be easily controlled => stability analysis
Waveform
modelling
In particular:
Moment
tensor
inversion
• good network configuration => focal sphere nicely covered
• inversion algorithm:
• optimal with same window length for Green’s functions and data
Summary
• optimal with simultaneous inversion of P- and S-wave, but excluding S-wave coda
• yields high isotropic percentage, higher than amplitude inversion, almost
independently of window length
promising, but computationally demanding (especially the computation of Green’s
functions with sufficiently small grid point distances)
will be performed for selected events, not whole database
Thank you
for your
attention!
http://commons.wikimedia.org/wiki/File:Preikestolen_Norge.jpg
Slide 21
Moment tensor inversion at
Pyhasalmi ore mine:
accuracy test using explosions
Daniela Kühn (NORSAR)
V. Vavrycuk (Academy of Sciences of the CR)
AIM 2nd annual meeting, 29-30 Sept 2011, Prague, Czech Republic
Pyhäsalmi ore mine, Finland
• microseismic monitoring:
since January 2003
safety of the underground
personnel
optimisation of mining
process
Introduction
Waveform
modelling
Moment
tensor
inversion
• network:
12 1-C geophones
+ 6 3-C geophones (ISS)
3-D geometry
sampling rate: < 3000 Hz
Summary
owned by Inmet Mining Co.
• events:
1500 events /months
(including blasting)
-2 < Mw < 1.5
Inconsistent polarities of P-wave first onset
Moment tensor inversion:
• homogeneous velocity model (as for locations)
• amplitude inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling
Complexity of velocity model
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling: 2D section
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
• E3D: viscoelastic 3-D FD code (Larsen and Schultz, 1995)
• strong interaction with mining cavities: reflection,
scattering, conversion
620 m
• healing of wavefronts
Waveform modelling
synthetic seismograms
- complex waveforms
Introduction
- strong coda
- complex secondary arrivals
Waveform
modelling
- scattering effects stronger
on amplitudes than travel
times,
since
size
of
heterogeneities
(cavities,
access tunnels) same order or
smaller than wavelengths
Moment
tensor
inversion
Summary
arrival times computed by
Eikonal solver still fit
(wavefronts heal quickly
after passing a cavitiy)
observed seismograms
Comparison 1-D/3-D
Imaginary
network!
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Comparison 1-D/3-D
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Influence of proximity to cavity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Real mine
network!
Source depth → ray path
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Ray path → onset polarity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Moment tensor inversion
Amplitude picking
direct wave scattered wave
Introduction
Waveform
modelling
Moment
tensor
inversion
first maximum amplitude = amplitude of the direct wave
direct wave
scattered wave
Summary
energy diffracted
around cavity
?
first maximum amplitude is not always the amplitude of the direct wave
Amplitude vs. waveform inversion
Amplitude inversion:
• homogeneous model of the medium
Introduction
• Green’s functions calculated using ray theory
• inversion of P-wave amplitudes (20-30 amplitudes)
Waveform
modelling
• frequencies: 250-500 Hz
cannot take into account distortion of rays on focal sphere
Moment
tensor
inversion
misinterpretation of amplitudes: which one is the direct wave?
Waveform inversion:
Summary
• 3-D heterogeneous model of the medium
• Green’s functions calculated using FD code
• inversion of full waveforms (15-20 waveforms)
• frequencies (at the moment): 25-100 Hz
inversion is performed in frequency domain
in principle same inversion algorithm as for amplitude inversion, but run
repeatedly for every frequency band (0.5 Hz steps)
Selected explosions: relocation
expl 1
expl 1
expl 3
Introduction
expl 5
Waveform
modelling
expl 3
Moment
tensor
inversion
expl 5
expl 1
Summary
expl 3
expl 5
Explosions (coords in m):
1) x=8306E y=2312N z=-1238
→ 26 m shift
3) x=8218E y=2192N z=-1352
→ 68 m shift
5) x=8214E y=2168N z=-1356
→ 59 m shift
Examples: good fit
Examples: amplitude misfit
Examples: phase misfit
Inversion results
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
explosion 1
explosion 3
explosion 5
Length of time windows
GF duration [10 ms]
Data duration [10 ms]
explosion 1
ISO percentage
GF duration [10 ms]
ISO percentage
GF duration [10 ms]
ISO percentage
Data duration [10 ms]
explosion 3
Data duration [10 ms]
explosion 5
nearly all solutions have high isotropic percentage
best solutions near diagonal
(length of GF time window = length of data time window)
best solutions P –wave + S-wave onset
Explosion 1
Amplitude inversion
Waveform inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
DC = 7%
CLVD = -14%
ISO = 79%
Summary: seismicity in mines
structural model in mines is very complex
Introduction
large and abrupt changes in velocity at
cavities
model varies in time
Waveform
modelling
Moment
tensor
inversion
Summary
earthquake source is complex (single
forces, non-DC components)
small changes in source position lead to
large changes in ray propagation, rays can
be strongly curved
radiated wave field is complex
(reflected, converted, scattered waves, head
waves)
Summary: waveform inversion
In general:
• complex Green’s functions can be calculated by 3-D FD codes (accurate model needed!);
Introduction
• sensitive to time shifts due to mislocation or due to inaccurate velocity model
• frequency band of inverted waves can be easily controlled => stability analysis
Waveform
modelling
In particular:
Moment
tensor
inversion
• good network configuration => focal sphere nicely covered
• inversion algorithm:
• optimal with same window length for Green’s functions and data
Summary
• optimal with simultaneous inversion of P- and S-wave, but excluding S-wave coda
• yields high isotropic percentage, higher than amplitude inversion, almost
independently of window length
promising, but computationally demanding (especially the computation of Green’s
functions with sufficiently small grid point distances)
will be performed for selected events, not whole database
Thank you
for your
attention!
http://commons.wikimedia.org/wiki/File:Preikestolen_Norge.jpg
Slide 22
Moment tensor inversion at
Pyhasalmi ore mine:
accuracy test using explosions
Daniela Kühn (NORSAR)
V. Vavrycuk (Academy of Sciences of the CR)
AIM 2nd annual meeting, 29-30 Sept 2011, Prague, Czech Republic
Pyhäsalmi ore mine, Finland
• microseismic monitoring:
since January 2003
safety of the underground
personnel
optimisation of mining
process
Introduction
Waveform
modelling
Moment
tensor
inversion
• network:
12 1-C geophones
+ 6 3-C geophones (ISS)
3-D geometry
sampling rate: < 3000 Hz
Summary
owned by Inmet Mining Co.
• events:
1500 events /months
(including blasting)
-2 < Mw < 1.5
Inconsistent polarities of P-wave first onset
Moment tensor inversion:
• homogeneous velocity model (as for locations)
• amplitude inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling
Complexity of velocity model
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling: 2D section
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
• E3D: viscoelastic 3-D FD code (Larsen and Schultz, 1995)
• strong interaction with mining cavities: reflection,
scattering, conversion
620 m
• healing of wavefronts
Waveform modelling
synthetic seismograms
- complex waveforms
Introduction
- strong coda
- complex secondary arrivals
Waveform
modelling
- scattering effects stronger
on amplitudes than travel
times,
since
size
of
heterogeneities
(cavities,
access tunnels) same order or
smaller than wavelengths
Moment
tensor
inversion
Summary
arrival times computed by
Eikonal solver still fit
(wavefronts heal quickly
after passing a cavitiy)
observed seismograms
Comparison 1-D/3-D
Imaginary
network!
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Comparison 1-D/3-D
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Influence of proximity to cavity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Real mine
network!
Source depth → ray path
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Ray path → onset polarity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Moment tensor inversion
Amplitude picking
direct wave scattered wave
Introduction
Waveform
modelling
Moment
tensor
inversion
first maximum amplitude = amplitude of the direct wave
direct wave
scattered wave
Summary
energy diffracted
around cavity
?
first maximum amplitude is not always the amplitude of the direct wave
Amplitude vs. waveform inversion
Amplitude inversion:
• homogeneous model of the medium
Introduction
• Green’s functions calculated using ray theory
• inversion of P-wave amplitudes (20-30 amplitudes)
Waveform
modelling
• frequencies: 250-500 Hz
cannot take into account distortion of rays on focal sphere
Moment
tensor
inversion
misinterpretation of amplitudes: which one is the direct wave?
Waveform inversion:
Summary
• 3-D heterogeneous model of the medium
• Green’s functions calculated using FD code
• inversion of full waveforms (15-20 waveforms)
• frequencies (at the moment): 25-100 Hz
inversion is performed in frequency domain
in principle same inversion algorithm as for amplitude inversion, but run
repeatedly for every frequency band (0.5 Hz steps)
Selected explosions: relocation
expl 1
expl 1
expl 3
Introduction
expl 5
Waveform
modelling
expl 3
Moment
tensor
inversion
expl 5
expl 1
Summary
expl 3
expl 5
Explosions (coords in m):
1) x=8306E y=2312N z=-1238
→ 26 m shift
3) x=8218E y=2192N z=-1352
→ 68 m shift
5) x=8214E y=2168N z=-1356
→ 59 m shift
Examples: good fit
Examples: amplitude misfit
Examples: phase misfit
Inversion results
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
explosion 1
explosion 3
explosion 5
Length of time windows
GF duration [10 ms]
Data duration [10 ms]
explosion 1
ISO percentage
GF duration [10 ms]
ISO percentage
GF duration [10 ms]
ISO percentage
Data duration [10 ms]
explosion 3
Data duration [10 ms]
explosion 5
nearly all solutions have high isotropic percentage
best solutions near diagonal
(length of GF time window = length of data time window)
best solutions P –wave + S-wave onset
Explosion 1
Amplitude inversion
Waveform inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
DC = 7%
CLVD = -14%
ISO = 79%
Summary: seismicity in mines
structural model in mines is very complex
Introduction
large and abrupt changes in velocity at
cavities
model varies in time
Waveform
modelling
Moment
tensor
inversion
Summary
earthquake source is complex (single
forces, non-DC components)
small changes in source position lead to
large changes in ray propagation, rays can
be strongly curved
radiated wave field is complex
(reflected, converted, scattered waves, head
waves)
Summary: waveform inversion
In general:
• complex Green’s functions can be calculated by 3-D FD codes (accurate model needed!);
Introduction
• sensitive to time shifts due to mislocation or due to inaccurate velocity model
• frequency band of inverted waves can be easily controlled => stability analysis
Waveform
modelling
In particular:
Moment
tensor
inversion
• good network configuration => focal sphere nicely covered
• inversion algorithm:
• optimal with same window length for Green’s functions and data
Summary
• optimal with simultaneous inversion of P- and S-wave, but excluding S-wave coda
• yields high isotropic percentage, higher than amplitude inversion, almost
independently of window length
promising, but computationally demanding (especially the computation of Green’s
functions with sufficiently small grid point distances)
will be performed for selected events, not whole database
Thank you
for your
attention!
http://commons.wikimedia.org/wiki/File:Preikestolen_Norge.jpg
Slide 23
Moment tensor inversion at
Pyhasalmi ore mine:
accuracy test using explosions
Daniela Kühn (NORSAR)
V. Vavrycuk (Academy of Sciences of the CR)
AIM 2nd annual meeting, 29-30 Sept 2011, Prague, Czech Republic
Pyhäsalmi ore mine, Finland
• microseismic monitoring:
since January 2003
safety of the underground
personnel
optimisation of mining
process
Introduction
Waveform
modelling
Moment
tensor
inversion
• network:
12 1-C geophones
+ 6 3-C geophones (ISS)
3-D geometry
sampling rate: < 3000 Hz
Summary
owned by Inmet Mining Co.
• events:
1500 events /months
(including blasting)
-2 < Mw < 1.5
Inconsistent polarities of P-wave first onset
Moment tensor inversion:
• homogeneous velocity model (as for locations)
• amplitude inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling
Complexity of velocity model
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling: 2D section
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
• E3D: viscoelastic 3-D FD code (Larsen and Schultz, 1995)
• strong interaction with mining cavities: reflection,
scattering, conversion
620 m
• healing of wavefronts
Waveform modelling
synthetic seismograms
- complex waveforms
Introduction
- strong coda
- complex secondary arrivals
Waveform
modelling
- scattering effects stronger
on amplitudes than travel
times,
since
size
of
heterogeneities
(cavities,
access tunnels) same order or
smaller than wavelengths
Moment
tensor
inversion
Summary
arrival times computed by
Eikonal solver still fit
(wavefronts heal quickly
after passing a cavitiy)
observed seismograms
Comparison 1-D/3-D
Imaginary
network!
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Comparison 1-D/3-D
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Influence of proximity to cavity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Real mine
network!
Source depth → ray path
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Ray path → onset polarity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Moment tensor inversion
Amplitude picking
direct wave scattered wave
Introduction
Waveform
modelling
Moment
tensor
inversion
first maximum amplitude = amplitude of the direct wave
direct wave
scattered wave
Summary
energy diffracted
around cavity
?
first maximum amplitude is not always the amplitude of the direct wave
Amplitude vs. waveform inversion
Amplitude inversion:
• homogeneous model of the medium
Introduction
• Green’s functions calculated using ray theory
• inversion of P-wave amplitudes (20-30 amplitudes)
Waveform
modelling
• frequencies: 250-500 Hz
cannot take into account distortion of rays on focal sphere
Moment
tensor
inversion
misinterpretation of amplitudes: which one is the direct wave?
Waveform inversion:
Summary
• 3-D heterogeneous model of the medium
• Green’s functions calculated using FD code
• inversion of full waveforms (15-20 waveforms)
• frequencies (at the moment): 25-100 Hz
inversion is performed in frequency domain
in principle same inversion algorithm as for amplitude inversion, but run
repeatedly for every frequency band (0.5 Hz steps)
Selected explosions: relocation
expl 1
expl 1
expl 3
Introduction
expl 5
Waveform
modelling
expl 3
Moment
tensor
inversion
expl 5
expl 1
Summary
expl 3
expl 5
Explosions (coords in m):
1) x=8306E y=2312N z=-1238
→ 26 m shift
3) x=8218E y=2192N z=-1352
→ 68 m shift
5) x=8214E y=2168N z=-1356
→ 59 m shift
Examples: good fit
Examples: amplitude misfit
Examples: phase misfit
Inversion results
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
explosion 1
explosion 3
explosion 5
Length of time windows
GF duration [10 ms]
Data duration [10 ms]
explosion 1
ISO percentage
GF duration [10 ms]
ISO percentage
GF duration [10 ms]
ISO percentage
Data duration [10 ms]
explosion 3
Data duration [10 ms]
explosion 5
nearly all solutions have high isotropic percentage
best solutions near diagonal
(length of GF time window = length of data time window)
best solutions P –wave + S-wave onset
Explosion 1
Amplitude inversion
Waveform inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
DC = 7%
CLVD = -14%
ISO = 79%
Summary: seismicity in mines
structural model in mines is very complex
Introduction
large and abrupt changes in velocity at
cavities
model varies in time
Waveform
modelling
Moment
tensor
inversion
Summary
earthquake source is complex (single
forces, non-DC components)
small changes in source position lead to
large changes in ray propagation, rays can
be strongly curved
radiated wave field is complex
(reflected, converted, scattered waves, head
waves)
Summary: waveform inversion
In general:
• complex Green’s functions can be calculated by 3-D FD codes (accurate model needed!);
Introduction
• sensitive to time shifts due to mislocation or due to inaccurate velocity model
• frequency band of inverted waves can be easily controlled => stability analysis
Waveform
modelling
In particular:
Moment
tensor
inversion
• good network configuration => focal sphere nicely covered
• inversion algorithm:
• optimal with same window length for Green’s functions and data
Summary
• optimal with simultaneous inversion of P- and S-wave, but excluding S-wave coda
• yields high isotropic percentage, higher than amplitude inversion, almost
independently of window length
promising, but computationally demanding (especially the computation of Green’s
functions with sufficiently small grid point distances)
will be performed for selected events, not whole database
Thank you
for your
attention!
http://commons.wikimedia.org/wiki/File:Preikestolen_Norge.jpg
Slide 24
Moment tensor inversion at
Pyhasalmi ore mine:
accuracy test using explosions
Daniela Kühn (NORSAR)
V. Vavrycuk (Academy of Sciences of the CR)
AIM 2nd annual meeting, 29-30 Sept 2011, Prague, Czech Republic
Pyhäsalmi ore mine, Finland
• microseismic monitoring:
since January 2003
safety of the underground
personnel
optimisation of mining
process
Introduction
Waveform
modelling
Moment
tensor
inversion
• network:
12 1-C geophones
+ 6 3-C geophones (ISS)
3-D geometry
sampling rate: < 3000 Hz
Summary
owned by Inmet Mining Co.
• events:
1500 events /months
(including blasting)
-2 < Mw < 1.5
Inconsistent polarities of P-wave first onset
Moment tensor inversion:
• homogeneous velocity model (as for locations)
• amplitude inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling
Complexity of velocity model
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling: 2D section
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
• E3D: viscoelastic 3-D FD code (Larsen and Schultz, 1995)
• strong interaction with mining cavities: reflection,
scattering, conversion
620 m
• healing of wavefronts
Waveform modelling
synthetic seismograms
- complex waveforms
Introduction
- strong coda
- complex secondary arrivals
Waveform
modelling
- scattering effects stronger
on amplitudes than travel
times,
since
size
of
heterogeneities
(cavities,
access tunnels) same order or
smaller than wavelengths
Moment
tensor
inversion
Summary
arrival times computed by
Eikonal solver still fit
(wavefronts heal quickly
after passing a cavitiy)
observed seismograms
Comparison 1-D/3-D
Imaginary
network!
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Comparison 1-D/3-D
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Influence of proximity to cavity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Real mine
network!
Source depth → ray path
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Ray path → onset polarity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Moment tensor inversion
Amplitude picking
direct wave scattered wave
Introduction
Waveform
modelling
Moment
tensor
inversion
first maximum amplitude = amplitude of the direct wave
direct wave
scattered wave
Summary
energy diffracted
around cavity
?
first maximum amplitude is not always the amplitude of the direct wave
Amplitude vs. waveform inversion
Amplitude inversion:
• homogeneous model of the medium
Introduction
• Green’s functions calculated using ray theory
• inversion of P-wave amplitudes (20-30 amplitudes)
Waveform
modelling
• frequencies: 250-500 Hz
cannot take into account distortion of rays on focal sphere
Moment
tensor
inversion
misinterpretation of amplitudes: which one is the direct wave?
Waveform inversion:
Summary
• 3-D heterogeneous model of the medium
• Green’s functions calculated using FD code
• inversion of full waveforms (15-20 waveforms)
• frequencies (at the moment): 25-100 Hz
inversion is performed in frequency domain
in principle same inversion algorithm as for amplitude inversion, but run
repeatedly for every frequency band (0.5 Hz steps)
Selected explosions: relocation
expl 1
expl 1
expl 3
Introduction
expl 5
Waveform
modelling
expl 3
Moment
tensor
inversion
expl 5
expl 1
Summary
expl 3
expl 5
Explosions (coords in m):
1) x=8306E y=2312N z=-1238
→ 26 m shift
3) x=8218E y=2192N z=-1352
→ 68 m shift
5) x=8214E y=2168N z=-1356
→ 59 m shift
Examples: good fit
Examples: amplitude misfit
Examples: phase misfit
Inversion results
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
explosion 1
explosion 3
explosion 5
Length of time windows
GF duration [10 ms]
Data duration [10 ms]
explosion 1
ISO percentage
GF duration [10 ms]
ISO percentage
GF duration [10 ms]
ISO percentage
Data duration [10 ms]
explosion 3
Data duration [10 ms]
explosion 5
nearly all solutions have high isotropic percentage
best solutions near diagonal
(length of GF time window = length of data time window)
best solutions P –wave + S-wave onset
Explosion 1
Amplitude inversion
Waveform inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
DC = 7%
CLVD = -14%
ISO = 79%
Summary: seismicity in mines
structural model in mines is very complex
Introduction
large and abrupt changes in velocity at
cavities
model varies in time
Waveform
modelling
Moment
tensor
inversion
Summary
earthquake source is complex (single
forces, non-DC components)
small changes in source position lead to
large changes in ray propagation, rays can
be strongly curved
radiated wave field is complex
(reflected, converted, scattered waves, head
waves)
Summary: waveform inversion
In general:
• complex Green’s functions can be calculated by 3-D FD codes (accurate model needed!);
Introduction
• sensitive to time shifts due to mislocation or due to inaccurate velocity model
• frequency band of inverted waves can be easily controlled => stability analysis
Waveform
modelling
In particular:
Moment
tensor
inversion
• good network configuration => focal sphere nicely covered
• inversion algorithm:
• optimal with same window length for Green’s functions and data
Summary
• optimal with simultaneous inversion of P- and S-wave, but excluding S-wave coda
• yields high isotropic percentage, higher than amplitude inversion, almost
independently of window length
promising, but computationally demanding (especially the computation of Green’s
functions with sufficiently small grid point distances)
will be performed for selected events, not whole database
Thank you
for your
attention!
http://commons.wikimedia.org/wiki/File:Preikestolen_Norge.jpg
Slide 25
Moment tensor inversion at
Pyhasalmi ore mine:
accuracy test using explosions
Daniela Kühn (NORSAR)
V. Vavrycuk (Academy of Sciences of the CR)
AIM 2nd annual meeting, 29-30 Sept 2011, Prague, Czech Republic
Pyhäsalmi ore mine, Finland
• microseismic monitoring:
since January 2003
safety of the underground
personnel
optimisation of mining
process
Introduction
Waveform
modelling
Moment
tensor
inversion
• network:
12 1-C geophones
+ 6 3-C geophones (ISS)
3-D geometry
sampling rate: < 3000 Hz
Summary
owned by Inmet Mining Co.
• events:
1500 events /months
(including blasting)
-2 < Mw < 1.5
Inconsistent polarities of P-wave first onset
Moment tensor inversion:
• homogeneous velocity model (as for locations)
• amplitude inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling
Complexity of velocity model
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling: 2D section
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
• E3D: viscoelastic 3-D FD code (Larsen and Schultz, 1995)
• strong interaction with mining cavities: reflection,
scattering, conversion
620 m
• healing of wavefronts
Waveform modelling
synthetic seismograms
- complex waveforms
Introduction
- strong coda
- complex secondary arrivals
Waveform
modelling
- scattering effects stronger
on amplitudes than travel
times,
since
size
of
heterogeneities
(cavities,
access tunnels) same order or
smaller than wavelengths
Moment
tensor
inversion
Summary
arrival times computed by
Eikonal solver still fit
(wavefronts heal quickly
after passing a cavitiy)
observed seismograms
Comparison 1-D/3-D
Imaginary
network!
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Comparison 1-D/3-D
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Influence of proximity to cavity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Real mine
network!
Source depth → ray path
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Ray path → onset polarity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Moment tensor inversion
Amplitude picking
direct wave scattered wave
Introduction
Waveform
modelling
Moment
tensor
inversion
first maximum amplitude = amplitude of the direct wave
direct wave
scattered wave
Summary
energy diffracted
around cavity
?
first maximum amplitude is not always the amplitude of the direct wave
Amplitude vs. waveform inversion
Amplitude inversion:
• homogeneous model of the medium
Introduction
• Green’s functions calculated using ray theory
• inversion of P-wave amplitudes (20-30 amplitudes)
Waveform
modelling
• frequencies: 250-500 Hz
cannot take into account distortion of rays on focal sphere
Moment
tensor
inversion
misinterpretation of amplitudes: which one is the direct wave?
Waveform inversion:
Summary
• 3-D heterogeneous model of the medium
• Green’s functions calculated using FD code
• inversion of full waveforms (15-20 waveforms)
• frequencies (at the moment): 25-100 Hz
inversion is performed in frequency domain
in principle same inversion algorithm as for amplitude inversion, but run
repeatedly for every frequency band (0.5 Hz steps)
Selected explosions: relocation
expl 1
expl 1
expl 3
Introduction
expl 5
Waveform
modelling
expl 3
Moment
tensor
inversion
expl 5
expl 1
Summary
expl 3
expl 5
Explosions (coords in m):
1) x=8306E y=2312N z=-1238
→ 26 m shift
3) x=8218E y=2192N z=-1352
→ 68 m shift
5) x=8214E y=2168N z=-1356
→ 59 m shift
Examples: good fit
Examples: amplitude misfit
Examples: phase misfit
Inversion results
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
explosion 1
explosion 3
explosion 5
Length of time windows
GF duration [10 ms]
Data duration [10 ms]
explosion 1
ISO percentage
GF duration [10 ms]
ISO percentage
GF duration [10 ms]
ISO percentage
Data duration [10 ms]
explosion 3
Data duration [10 ms]
explosion 5
nearly all solutions have high isotropic percentage
best solutions near diagonal
(length of GF time window = length of data time window)
best solutions P –wave + S-wave onset
Explosion 1
Amplitude inversion
Waveform inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
DC = 7%
CLVD = -14%
ISO = 79%
Summary: seismicity in mines
structural model in mines is very complex
Introduction
large and abrupt changes in velocity at
cavities
model varies in time
Waveform
modelling
Moment
tensor
inversion
Summary
earthquake source is complex (single
forces, non-DC components)
small changes in source position lead to
large changes in ray propagation, rays can
be strongly curved
radiated wave field is complex
(reflected, converted, scattered waves, head
waves)
Summary: waveform inversion
In general:
• complex Green’s functions can be calculated by 3-D FD codes (accurate model needed!);
Introduction
• sensitive to time shifts due to mislocation or due to inaccurate velocity model
• frequency band of inverted waves can be easily controlled => stability analysis
Waveform
modelling
In particular:
Moment
tensor
inversion
• good network configuration => focal sphere nicely covered
• inversion algorithm:
• optimal with same window length for Green’s functions and data
Summary
• optimal with simultaneous inversion of P- and S-wave, but excluding S-wave coda
• yields high isotropic percentage, higher than amplitude inversion, almost
independently of window length
promising, but computationally demanding (especially the computation of Green’s
functions with sufficiently small grid point distances)
will be performed for selected events, not whole database
Thank you
for your
attention!
http://commons.wikimedia.org/wiki/File:Preikestolen_Norge.jpg
Moment tensor inversion at
Pyhasalmi ore mine:
accuracy test using explosions
Daniela Kühn (NORSAR)
V. Vavrycuk (Academy of Sciences of the CR)
AIM 2nd annual meeting, 29-30 Sept 2011, Prague, Czech Republic
Pyhäsalmi ore mine, Finland
• microseismic monitoring:
since January 2003
safety of the underground
personnel
optimisation of mining
process
Introduction
Waveform
modelling
Moment
tensor
inversion
• network:
12 1-C geophones
+ 6 3-C geophones (ISS)
3-D geometry
sampling rate: < 3000 Hz
Summary
owned by Inmet Mining Co.
• events:
1500 events /months
(including blasting)
-2 < Mw < 1.5
Inconsistent polarities of P-wave first onset
Moment tensor inversion:
• homogeneous velocity model (as for locations)
• amplitude inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling
Complexity of velocity model
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling: 2D section
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
• E3D: viscoelastic 3-D FD code (Larsen and Schultz, 1995)
• strong interaction with mining cavities: reflection,
scattering, conversion
620 m
• healing of wavefronts
Waveform modelling
synthetic seismograms
- complex waveforms
Introduction
- strong coda
- complex secondary arrivals
Waveform
modelling
- scattering effects stronger
on amplitudes than travel
times,
since
size
of
heterogeneities
(cavities,
access tunnels) same order or
smaller than wavelengths
Moment
tensor
inversion
Summary
arrival times computed by
Eikonal solver still fit
(wavefronts heal quickly
after passing a cavitiy)
observed seismograms
Comparison 1-D/3-D
Imaginary
network!
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Comparison 1-D/3-D
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Influence of proximity to cavity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Real mine
network!
Source depth → ray path
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Ray path → onset polarity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Moment tensor inversion
Amplitude picking
direct wave scattered wave
Introduction
Waveform
modelling
Moment
tensor
inversion
first maximum amplitude = amplitude of the direct wave
direct wave
scattered wave
Summary
energy diffracted
around cavity
?
first maximum amplitude is not always the amplitude of the direct wave
Amplitude vs. waveform inversion
Amplitude inversion:
• homogeneous model of the medium
Introduction
• Green’s functions calculated using ray theory
• inversion of P-wave amplitudes (20-30 amplitudes)
Waveform
modelling
• frequencies: 250-500 Hz
cannot take into account distortion of rays on focal sphere
Moment
tensor
inversion
misinterpretation of amplitudes: which one is the direct wave?
Waveform inversion:
Summary
• 3-D heterogeneous model of the medium
• Green’s functions calculated using FD code
• inversion of full waveforms (15-20 waveforms)
• frequencies (at the moment): 25-100 Hz
inversion is performed in frequency domain
in principle same inversion algorithm as for amplitude inversion, but run
repeatedly for every frequency band (0.5 Hz steps)
Selected explosions: relocation
expl 1
expl 1
expl 3
Introduction
expl 5
Waveform
modelling
expl 3
Moment
tensor
inversion
expl 5
expl 1
Summary
expl 3
expl 5
Explosions (coords in m):
1) x=8306E y=2312N z=-1238
→ 26 m shift
3) x=8218E y=2192N z=-1352
→ 68 m shift
5) x=8214E y=2168N z=-1356
→ 59 m shift
Examples: good fit
Examples: amplitude misfit
Examples: phase misfit
Inversion results
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
explosion 1
explosion 3
explosion 5
Length of time windows
GF duration [10 ms]
Data duration [10 ms]
explosion 1
ISO percentage
GF duration [10 ms]
ISO percentage
GF duration [10 ms]
ISO percentage
Data duration [10 ms]
explosion 3
Data duration [10 ms]
explosion 5
nearly all solutions have high isotropic percentage
best solutions near diagonal
(length of GF time window = length of data time window)
best solutions P –wave + S-wave onset
Explosion 1
Amplitude inversion
Waveform inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
DC = 7%
CLVD = -14%
ISO = 79%
Summary: seismicity in mines
structural model in mines is very complex
Introduction
large and abrupt changes in velocity at
cavities
model varies in time
Waveform
modelling
Moment
tensor
inversion
Summary
earthquake source is complex (single
forces, non-DC components)
small changes in source position lead to
large changes in ray propagation, rays can
be strongly curved
radiated wave field is complex
(reflected, converted, scattered waves, head
waves)
Summary: waveform inversion
In general:
• complex Green’s functions can be calculated by 3-D FD codes (accurate model needed!);
Introduction
• sensitive to time shifts due to mislocation or due to inaccurate velocity model
• frequency band of inverted waves can be easily controlled => stability analysis
Waveform
modelling
In particular:
Moment
tensor
inversion
• good network configuration => focal sphere nicely covered
• inversion algorithm:
• optimal with same window length for Green’s functions and data
Summary
• optimal with simultaneous inversion of P- and S-wave, but excluding S-wave coda
• yields high isotropic percentage, higher than amplitude inversion, almost
independently of window length
promising, but computationally demanding (especially the computation of Green’s
functions with sufficiently small grid point distances)
will be performed for selected events, not whole database
Thank you
for your
attention!
http://commons.wikimedia.org/wiki/File:Preikestolen_Norge.jpg
Slide 2
Moment tensor inversion at
Pyhasalmi ore mine:
accuracy test using explosions
Daniela Kühn (NORSAR)
V. Vavrycuk (Academy of Sciences of the CR)
AIM 2nd annual meeting, 29-30 Sept 2011, Prague, Czech Republic
Pyhäsalmi ore mine, Finland
• microseismic monitoring:
since January 2003
safety of the underground
personnel
optimisation of mining
process
Introduction
Waveform
modelling
Moment
tensor
inversion
• network:
12 1-C geophones
+ 6 3-C geophones (ISS)
3-D geometry
sampling rate: < 3000 Hz
Summary
owned by Inmet Mining Co.
• events:
1500 events /months
(including blasting)
-2 < Mw < 1.5
Inconsistent polarities of P-wave first onset
Moment tensor inversion:
• homogeneous velocity model (as for locations)
• amplitude inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling
Complexity of velocity model
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling: 2D section
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
• E3D: viscoelastic 3-D FD code (Larsen and Schultz, 1995)
• strong interaction with mining cavities: reflection,
scattering, conversion
620 m
• healing of wavefronts
Waveform modelling
synthetic seismograms
- complex waveforms
Introduction
- strong coda
- complex secondary arrivals
Waveform
modelling
- scattering effects stronger
on amplitudes than travel
times,
since
size
of
heterogeneities
(cavities,
access tunnels) same order or
smaller than wavelengths
Moment
tensor
inversion
Summary
arrival times computed by
Eikonal solver still fit
(wavefronts heal quickly
after passing a cavitiy)
observed seismograms
Comparison 1-D/3-D
Imaginary
network!
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Comparison 1-D/3-D
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Influence of proximity to cavity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Real mine
network!
Source depth → ray path
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Ray path → onset polarity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Moment tensor inversion
Amplitude picking
direct wave scattered wave
Introduction
Waveform
modelling
Moment
tensor
inversion
first maximum amplitude = amplitude of the direct wave
direct wave
scattered wave
Summary
energy diffracted
around cavity
?
first maximum amplitude is not always the amplitude of the direct wave
Amplitude vs. waveform inversion
Amplitude inversion:
• homogeneous model of the medium
Introduction
• Green’s functions calculated using ray theory
• inversion of P-wave amplitudes (20-30 amplitudes)
Waveform
modelling
• frequencies: 250-500 Hz
cannot take into account distortion of rays on focal sphere
Moment
tensor
inversion
misinterpretation of amplitudes: which one is the direct wave?
Waveform inversion:
Summary
• 3-D heterogeneous model of the medium
• Green’s functions calculated using FD code
• inversion of full waveforms (15-20 waveforms)
• frequencies (at the moment): 25-100 Hz
inversion is performed in frequency domain
in principle same inversion algorithm as for amplitude inversion, but run
repeatedly for every frequency band (0.5 Hz steps)
Selected explosions: relocation
expl 1
expl 1
expl 3
Introduction
expl 5
Waveform
modelling
expl 3
Moment
tensor
inversion
expl 5
expl 1
Summary
expl 3
expl 5
Explosions (coords in m):
1) x=8306E y=2312N z=-1238
→ 26 m shift
3) x=8218E y=2192N z=-1352
→ 68 m shift
5) x=8214E y=2168N z=-1356
→ 59 m shift
Examples: good fit
Examples: amplitude misfit
Examples: phase misfit
Inversion results
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
explosion 1
explosion 3
explosion 5
Length of time windows
GF duration [10 ms]
Data duration [10 ms]
explosion 1
ISO percentage
GF duration [10 ms]
ISO percentage
GF duration [10 ms]
ISO percentage
Data duration [10 ms]
explosion 3
Data duration [10 ms]
explosion 5
nearly all solutions have high isotropic percentage
best solutions near diagonal
(length of GF time window = length of data time window)
best solutions P –wave + S-wave onset
Explosion 1
Amplitude inversion
Waveform inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
DC = 7%
CLVD = -14%
ISO = 79%
Summary: seismicity in mines
structural model in mines is very complex
Introduction
large and abrupt changes in velocity at
cavities
model varies in time
Waveform
modelling
Moment
tensor
inversion
Summary
earthquake source is complex (single
forces, non-DC components)
small changes in source position lead to
large changes in ray propagation, rays can
be strongly curved
radiated wave field is complex
(reflected, converted, scattered waves, head
waves)
Summary: waveform inversion
In general:
• complex Green’s functions can be calculated by 3-D FD codes (accurate model needed!);
Introduction
• sensitive to time shifts due to mislocation or due to inaccurate velocity model
• frequency band of inverted waves can be easily controlled => stability analysis
Waveform
modelling
In particular:
Moment
tensor
inversion
• good network configuration => focal sphere nicely covered
• inversion algorithm:
• optimal with same window length for Green’s functions and data
Summary
• optimal with simultaneous inversion of P- and S-wave, but excluding S-wave coda
• yields high isotropic percentage, higher than amplitude inversion, almost
independently of window length
promising, but computationally demanding (especially the computation of Green’s
functions with sufficiently small grid point distances)
will be performed for selected events, not whole database
Thank you
for your
attention!
http://commons.wikimedia.org/wiki/File:Preikestolen_Norge.jpg
Slide 3
Moment tensor inversion at
Pyhasalmi ore mine:
accuracy test using explosions
Daniela Kühn (NORSAR)
V. Vavrycuk (Academy of Sciences of the CR)
AIM 2nd annual meeting, 29-30 Sept 2011, Prague, Czech Republic
Pyhäsalmi ore mine, Finland
• microseismic monitoring:
since January 2003
safety of the underground
personnel
optimisation of mining
process
Introduction
Waveform
modelling
Moment
tensor
inversion
• network:
12 1-C geophones
+ 6 3-C geophones (ISS)
3-D geometry
sampling rate: < 3000 Hz
Summary
owned by Inmet Mining Co.
• events:
1500 events /months
(including blasting)
-2 < Mw < 1.5
Inconsistent polarities of P-wave first onset
Moment tensor inversion:
• homogeneous velocity model (as for locations)
• amplitude inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling
Complexity of velocity model
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling: 2D section
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
• E3D: viscoelastic 3-D FD code (Larsen and Schultz, 1995)
• strong interaction with mining cavities: reflection,
scattering, conversion
620 m
• healing of wavefronts
Waveform modelling
synthetic seismograms
- complex waveforms
Introduction
- strong coda
- complex secondary arrivals
Waveform
modelling
- scattering effects stronger
on amplitudes than travel
times,
since
size
of
heterogeneities
(cavities,
access tunnels) same order or
smaller than wavelengths
Moment
tensor
inversion
Summary
arrival times computed by
Eikonal solver still fit
(wavefronts heal quickly
after passing a cavitiy)
observed seismograms
Comparison 1-D/3-D
Imaginary
network!
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Comparison 1-D/3-D
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Influence of proximity to cavity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Real mine
network!
Source depth → ray path
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Ray path → onset polarity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Moment tensor inversion
Amplitude picking
direct wave scattered wave
Introduction
Waveform
modelling
Moment
tensor
inversion
first maximum amplitude = amplitude of the direct wave
direct wave
scattered wave
Summary
energy diffracted
around cavity
?
first maximum amplitude is not always the amplitude of the direct wave
Amplitude vs. waveform inversion
Amplitude inversion:
• homogeneous model of the medium
Introduction
• Green’s functions calculated using ray theory
• inversion of P-wave amplitudes (20-30 amplitudes)
Waveform
modelling
• frequencies: 250-500 Hz
cannot take into account distortion of rays on focal sphere
Moment
tensor
inversion
misinterpretation of amplitudes: which one is the direct wave?
Waveform inversion:
Summary
• 3-D heterogeneous model of the medium
• Green’s functions calculated using FD code
• inversion of full waveforms (15-20 waveforms)
• frequencies (at the moment): 25-100 Hz
inversion is performed in frequency domain
in principle same inversion algorithm as for amplitude inversion, but run
repeatedly for every frequency band (0.5 Hz steps)
Selected explosions: relocation
expl 1
expl 1
expl 3
Introduction
expl 5
Waveform
modelling
expl 3
Moment
tensor
inversion
expl 5
expl 1
Summary
expl 3
expl 5
Explosions (coords in m):
1) x=8306E y=2312N z=-1238
→ 26 m shift
3) x=8218E y=2192N z=-1352
→ 68 m shift
5) x=8214E y=2168N z=-1356
→ 59 m shift
Examples: good fit
Examples: amplitude misfit
Examples: phase misfit
Inversion results
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
explosion 1
explosion 3
explosion 5
Length of time windows
GF duration [10 ms]
Data duration [10 ms]
explosion 1
ISO percentage
GF duration [10 ms]
ISO percentage
GF duration [10 ms]
ISO percentage
Data duration [10 ms]
explosion 3
Data duration [10 ms]
explosion 5
nearly all solutions have high isotropic percentage
best solutions near diagonal
(length of GF time window = length of data time window)
best solutions P –wave + S-wave onset
Explosion 1
Amplitude inversion
Waveform inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
DC = 7%
CLVD = -14%
ISO = 79%
Summary: seismicity in mines
structural model in mines is very complex
Introduction
large and abrupt changes in velocity at
cavities
model varies in time
Waveform
modelling
Moment
tensor
inversion
Summary
earthquake source is complex (single
forces, non-DC components)
small changes in source position lead to
large changes in ray propagation, rays can
be strongly curved
radiated wave field is complex
(reflected, converted, scattered waves, head
waves)
Summary: waveform inversion
In general:
• complex Green’s functions can be calculated by 3-D FD codes (accurate model needed!);
Introduction
• sensitive to time shifts due to mislocation or due to inaccurate velocity model
• frequency band of inverted waves can be easily controlled => stability analysis
Waveform
modelling
In particular:
Moment
tensor
inversion
• good network configuration => focal sphere nicely covered
• inversion algorithm:
• optimal with same window length for Green’s functions and data
Summary
• optimal with simultaneous inversion of P- and S-wave, but excluding S-wave coda
• yields high isotropic percentage, higher than amplitude inversion, almost
independently of window length
promising, but computationally demanding (especially the computation of Green’s
functions with sufficiently small grid point distances)
will be performed for selected events, not whole database
Thank you
for your
attention!
http://commons.wikimedia.org/wiki/File:Preikestolen_Norge.jpg
Slide 4
Moment tensor inversion at
Pyhasalmi ore mine:
accuracy test using explosions
Daniela Kühn (NORSAR)
V. Vavrycuk (Academy of Sciences of the CR)
AIM 2nd annual meeting, 29-30 Sept 2011, Prague, Czech Republic
Pyhäsalmi ore mine, Finland
• microseismic monitoring:
since January 2003
safety of the underground
personnel
optimisation of mining
process
Introduction
Waveform
modelling
Moment
tensor
inversion
• network:
12 1-C geophones
+ 6 3-C geophones (ISS)
3-D geometry
sampling rate: < 3000 Hz
Summary
owned by Inmet Mining Co.
• events:
1500 events /months
(including blasting)
-2 < Mw < 1.5
Inconsistent polarities of P-wave first onset
Moment tensor inversion:
• homogeneous velocity model (as for locations)
• amplitude inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling
Complexity of velocity model
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling: 2D section
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
• E3D: viscoelastic 3-D FD code (Larsen and Schultz, 1995)
• strong interaction with mining cavities: reflection,
scattering, conversion
620 m
• healing of wavefronts
Waveform modelling
synthetic seismograms
- complex waveforms
Introduction
- strong coda
- complex secondary arrivals
Waveform
modelling
- scattering effects stronger
on amplitudes than travel
times,
since
size
of
heterogeneities
(cavities,
access tunnels) same order or
smaller than wavelengths
Moment
tensor
inversion
Summary
arrival times computed by
Eikonal solver still fit
(wavefronts heal quickly
after passing a cavitiy)
observed seismograms
Comparison 1-D/3-D
Imaginary
network!
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Comparison 1-D/3-D
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Influence of proximity to cavity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Real mine
network!
Source depth → ray path
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Ray path → onset polarity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Moment tensor inversion
Amplitude picking
direct wave scattered wave
Introduction
Waveform
modelling
Moment
tensor
inversion
first maximum amplitude = amplitude of the direct wave
direct wave
scattered wave
Summary
energy diffracted
around cavity
?
first maximum amplitude is not always the amplitude of the direct wave
Amplitude vs. waveform inversion
Amplitude inversion:
• homogeneous model of the medium
Introduction
• Green’s functions calculated using ray theory
• inversion of P-wave amplitudes (20-30 amplitudes)
Waveform
modelling
• frequencies: 250-500 Hz
cannot take into account distortion of rays on focal sphere
Moment
tensor
inversion
misinterpretation of amplitudes: which one is the direct wave?
Waveform inversion:
Summary
• 3-D heterogeneous model of the medium
• Green’s functions calculated using FD code
• inversion of full waveforms (15-20 waveforms)
• frequencies (at the moment): 25-100 Hz
inversion is performed in frequency domain
in principle same inversion algorithm as for amplitude inversion, but run
repeatedly for every frequency band (0.5 Hz steps)
Selected explosions: relocation
expl 1
expl 1
expl 3
Introduction
expl 5
Waveform
modelling
expl 3
Moment
tensor
inversion
expl 5
expl 1
Summary
expl 3
expl 5
Explosions (coords in m):
1) x=8306E y=2312N z=-1238
→ 26 m shift
3) x=8218E y=2192N z=-1352
→ 68 m shift
5) x=8214E y=2168N z=-1356
→ 59 m shift
Examples: good fit
Examples: amplitude misfit
Examples: phase misfit
Inversion results
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
explosion 1
explosion 3
explosion 5
Length of time windows
GF duration [10 ms]
Data duration [10 ms]
explosion 1
ISO percentage
GF duration [10 ms]
ISO percentage
GF duration [10 ms]
ISO percentage
Data duration [10 ms]
explosion 3
Data duration [10 ms]
explosion 5
nearly all solutions have high isotropic percentage
best solutions near diagonal
(length of GF time window = length of data time window)
best solutions P –wave + S-wave onset
Explosion 1
Amplitude inversion
Waveform inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
DC = 7%
CLVD = -14%
ISO = 79%
Summary: seismicity in mines
structural model in mines is very complex
Introduction
large and abrupt changes in velocity at
cavities
model varies in time
Waveform
modelling
Moment
tensor
inversion
Summary
earthquake source is complex (single
forces, non-DC components)
small changes in source position lead to
large changes in ray propagation, rays can
be strongly curved
radiated wave field is complex
(reflected, converted, scattered waves, head
waves)
Summary: waveform inversion
In general:
• complex Green’s functions can be calculated by 3-D FD codes (accurate model needed!);
Introduction
• sensitive to time shifts due to mislocation or due to inaccurate velocity model
• frequency band of inverted waves can be easily controlled => stability analysis
Waveform
modelling
In particular:
Moment
tensor
inversion
• good network configuration => focal sphere nicely covered
• inversion algorithm:
• optimal with same window length for Green’s functions and data
Summary
• optimal with simultaneous inversion of P- and S-wave, but excluding S-wave coda
• yields high isotropic percentage, higher than amplitude inversion, almost
independently of window length
promising, but computationally demanding (especially the computation of Green’s
functions with sufficiently small grid point distances)
will be performed for selected events, not whole database
Thank you
for your
attention!
http://commons.wikimedia.org/wiki/File:Preikestolen_Norge.jpg
Slide 5
Moment tensor inversion at
Pyhasalmi ore mine:
accuracy test using explosions
Daniela Kühn (NORSAR)
V. Vavrycuk (Academy of Sciences of the CR)
AIM 2nd annual meeting, 29-30 Sept 2011, Prague, Czech Republic
Pyhäsalmi ore mine, Finland
• microseismic monitoring:
since January 2003
safety of the underground
personnel
optimisation of mining
process
Introduction
Waveform
modelling
Moment
tensor
inversion
• network:
12 1-C geophones
+ 6 3-C geophones (ISS)
3-D geometry
sampling rate: < 3000 Hz
Summary
owned by Inmet Mining Co.
• events:
1500 events /months
(including blasting)
-2 < Mw < 1.5
Inconsistent polarities of P-wave first onset
Moment tensor inversion:
• homogeneous velocity model (as for locations)
• amplitude inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling
Complexity of velocity model
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling: 2D section
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
• E3D: viscoelastic 3-D FD code (Larsen and Schultz, 1995)
• strong interaction with mining cavities: reflection,
scattering, conversion
620 m
• healing of wavefronts
Waveform modelling
synthetic seismograms
- complex waveforms
Introduction
- strong coda
- complex secondary arrivals
Waveform
modelling
- scattering effects stronger
on amplitudes than travel
times,
since
size
of
heterogeneities
(cavities,
access tunnels) same order or
smaller than wavelengths
Moment
tensor
inversion
Summary
arrival times computed by
Eikonal solver still fit
(wavefronts heal quickly
after passing a cavitiy)
observed seismograms
Comparison 1-D/3-D
Imaginary
network!
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Comparison 1-D/3-D
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Influence of proximity to cavity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Real mine
network!
Source depth → ray path
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Ray path → onset polarity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Moment tensor inversion
Amplitude picking
direct wave scattered wave
Introduction
Waveform
modelling
Moment
tensor
inversion
first maximum amplitude = amplitude of the direct wave
direct wave
scattered wave
Summary
energy diffracted
around cavity
?
first maximum amplitude is not always the amplitude of the direct wave
Amplitude vs. waveform inversion
Amplitude inversion:
• homogeneous model of the medium
Introduction
• Green’s functions calculated using ray theory
• inversion of P-wave amplitudes (20-30 amplitudes)
Waveform
modelling
• frequencies: 250-500 Hz
cannot take into account distortion of rays on focal sphere
Moment
tensor
inversion
misinterpretation of amplitudes: which one is the direct wave?
Waveform inversion:
Summary
• 3-D heterogeneous model of the medium
• Green’s functions calculated using FD code
• inversion of full waveforms (15-20 waveforms)
• frequencies (at the moment): 25-100 Hz
inversion is performed in frequency domain
in principle same inversion algorithm as for amplitude inversion, but run
repeatedly for every frequency band (0.5 Hz steps)
Selected explosions: relocation
expl 1
expl 1
expl 3
Introduction
expl 5
Waveform
modelling
expl 3
Moment
tensor
inversion
expl 5
expl 1
Summary
expl 3
expl 5
Explosions (coords in m):
1) x=8306E y=2312N z=-1238
→ 26 m shift
3) x=8218E y=2192N z=-1352
→ 68 m shift
5) x=8214E y=2168N z=-1356
→ 59 m shift
Examples: good fit
Examples: amplitude misfit
Examples: phase misfit
Inversion results
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
explosion 1
explosion 3
explosion 5
Length of time windows
GF duration [10 ms]
Data duration [10 ms]
explosion 1
ISO percentage
GF duration [10 ms]
ISO percentage
GF duration [10 ms]
ISO percentage
Data duration [10 ms]
explosion 3
Data duration [10 ms]
explosion 5
nearly all solutions have high isotropic percentage
best solutions near diagonal
(length of GF time window = length of data time window)
best solutions P –wave + S-wave onset
Explosion 1
Amplitude inversion
Waveform inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
DC = 7%
CLVD = -14%
ISO = 79%
Summary: seismicity in mines
structural model in mines is very complex
Introduction
large and abrupt changes in velocity at
cavities
model varies in time
Waveform
modelling
Moment
tensor
inversion
Summary
earthquake source is complex (single
forces, non-DC components)
small changes in source position lead to
large changes in ray propagation, rays can
be strongly curved
radiated wave field is complex
(reflected, converted, scattered waves, head
waves)
Summary: waveform inversion
In general:
• complex Green’s functions can be calculated by 3-D FD codes (accurate model needed!);
Introduction
• sensitive to time shifts due to mislocation or due to inaccurate velocity model
• frequency band of inverted waves can be easily controlled => stability analysis
Waveform
modelling
In particular:
Moment
tensor
inversion
• good network configuration => focal sphere nicely covered
• inversion algorithm:
• optimal with same window length for Green’s functions and data
Summary
• optimal with simultaneous inversion of P- and S-wave, but excluding S-wave coda
• yields high isotropic percentage, higher than amplitude inversion, almost
independently of window length
promising, but computationally demanding (especially the computation of Green’s
functions with sufficiently small grid point distances)
will be performed for selected events, not whole database
Thank you
for your
attention!
http://commons.wikimedia.org/wiki/File:Preikestolen_Norge.jpg
Slide 6
Moment tensor inversion at
Pyhasalmi ore mine:
accuracy test using explosions
Daniela Kühn (NORSAR)
V. Vavrycuk (Academy of Sciences of the CR)
AIM 2nd annual meeting, 29-30 Sept 2011, Prague, Czech Republic
Pyhäsalmi ore mine, Finland
• microseismic monitoring:
since January 2003
safety of the underground
personnel
optimisation of mining
process
Introduction
Waveform
modelling
Moment
tensor
inversion
• network:
12 1-C geophones
+ 6 3-C geophones (ISS)
3-D geometry
sampling rate: < 3000 Hz
Summary
owned by Inmet Mining Co.
• events:
1500 events /months
(including blasting)
-2 < Mw < 1.5
Inconsistent polarities of P-wave first onset
Moment tensor inversion:
• homogeneous velocity model (as for locations)
• amplitude inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling
Complexity of velocity model
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling: 2D section
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
• E3D: viscoelastic 3-D FD code (Larsen and Schultz, 1995)
• strong interaction with mining cavities: reflection,
scattering, conversion
620 m
• healing of wavefronts
Waveform modelling
synthetic seismograms
- complex waveforms
Introduction
- strong coda
- complex secondary arrivals
Waveform
modelling
- scattering effects stronger
on amplitudes than travel
times,
since
size
of
heterogeneities
(cavities,
access tunnels) same order or
smaller than wavelengths
Moment
tensor
inversion
Summary
arrival times computed by
Eikonal solver still fit
(wavefronts heal quickly
after passing a cavitiy)
observed seismograms
Comparison 1-D/3-D
Imaginary
network!
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Comparison 1-D/3-D
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Influence of proximity to cavity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Real mine
network!
Source depth → ray path
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Ray path → onset polarity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Moment tensor inversion
Amplitude picking
direct wave scattered wave
Introduction
Waveform
modelling
Moment
tensor
inversion
first maximum amplitude = amplitude of the direct wave
direct wave
scattered wave
Summary
energy diffracted
around cavity
?
first maximum amplitude is not always the amplitude of the direct wave
Amplitude vs. waveform inversion
Amplitude inversion:
• homogeneous model of the medium
Introduction
• Green’s functions calculated using ray theory
• inversion of P-wave amplitudes (20-30 amplitudes)
Waveform
modelling
• frequencies: 250-500 Hz
cannot take into account distortion of rays on focal sphere
Moment
tensor
inversion
misinterpretation of amplitudes: which one is the direct wave?
Waveform inversion:
Summary
• 3-D heterogeneous model of the medium
• Green’s functions calculated using FD code
• inversion of full waveforms (15-20 waveforms)
• frequencies (at the moment): 25-100 Hz
inversion is performed in frequency domain
in principle same inversion algorithm as for amplitude inversion, but run
repeatedly for every frequency band (0.5 Hz steps)
Selected explosions: relocation
expl 1
expl 1
expl 3
Introduction
expl 5
Waveform
modelling
expl 3
Moment
tensor
inversion
expl 5
expl 1
Summary
expl 3
expl 5
Explosions (coords in m):
1) x=8306E y=2312N z=-1238
→ 26 m shift
3) x=8218E y=2192N z=-1352
→ 68 m shift
5) x=8214E y=2168N z=-1356
→ 59 m shift
Examples: good fit
Examples: amplitude misfit
Examples: phase misfit
Inversion results
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
explosion 1
explosion 3
explosion 5
Length of time windows
GF duration [10 ms]
Data duration [10 ms]
explosion 1
ISO percentage
GF duration [10 ms]
ISO percentage
GF duration [10 ms]
ISO percentage
Data duration [10 ms]
explosion 3
Data duration [10 ms]
explosion 5
nearly all solutions have high isotropic percentage
best solutions near diagonal
(length of GF time window = length of data time window)
best solutions P –wave + S-wave onset
Explosion 1
Amplitude inversion
Waveform inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
DC = 7%
CLVD = -14%
ISO = 79%
Summary: seismicity in mines
structural model in mines is very complex
Introduction
large and abrupt changes in velocity at
cavities
model varies in time
Waveform
modelling
Moment
tensor
inversion
Summary
earthquake source is complex (single
forces, non-DC components)
small changes in source position lead to
large changes in ray propagation, rays can
be strongly curved
radiated wave field is complex
(reflected, converted, scattered waves, head
waves)
Summary: waveform inversion
In general:
• complex Green’s functions can be calculated by 3-D FD codes (accurate model needed!);
Introduction
• sensitive to time shifts due to mislocation or due to inaccurate velocity model
• frequency band of inverted waves can be easily controlled => stability analysis
Waveform
modelling
In particular:
Moment
tensor
inversion
• good network configuration => focal sphere nicely covered
• inversion algorithm:
• optimal with same window length for Green’s functions and data
Summary
• optimal with simultaneous inversion of P- and S-wave, but excluding S-wave coda
• yields high isotropic percentage, higher than amplitude inversion, almost
independently of window length
promising, but computationally demanding (especially the computation of Green’s
functions with sufficiently small grid point distances)
will be performed for selected events, not whole database
Thank you
for your
attention!
http://commons.wikimedia.org/wiki/File:Preikestolen_Norge.jpg
Slide 7
Moment tensor inversion at
Pyhasalmi ore mine:
accuracy test using explosions
Daniela Kühn (NORSAR)
V. Vavrycuk (Academy of Sciences of the CR)
AIM 2nd annual meeting, 29-30 Sept 2011, Prague, Czech Republic
Pyhäsalmi ore mine, Finland
• microseismic monitoring:
since January 2003
safety of the underground
personnel
optimisation of mining
process
Introduction
Waveform
modelling
Moment
tensor
inversion
• network:
12 1-C geophones
+ 6 3-C geophones (ISS)
3-D geometry
sampling rate: < 3000 Hz
Summary
owned by Inmet Mining Co.
• events:
1500 events /months
(including blasting)
-2 < Mw < 1.5
Inconsistent polarities of P-wave first onset
Moment tensor inversion:
• homogeneous velocity model (as for locations)
• amplitude inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling
Complexity of velocity model
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling: 2D section
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
• E3D: viscoelastic 3-D FD code (Larsen and Schultz, 1995)
• strong interaction with mining cavities: reflection,
scattering, conversion
620 m
• healing of wavefronts
Waveform modelling
synthetic seismograms
- complex waveforms
Introduction
- strong coda
- complex secondary arrivals
Waveform
modelling
- scattering effects stronger
on amplitudes than travel
times,
since
size
of
heterogeneities
(cavities,
access tunnels) same order or
smaller than wavelengths
Moment
tensor
inversion
Summary
arrival times computed by
Eikonal solver still fit
(wavefronts heal quickly
after passing a cavitiy)
observed seismograms
Comparison 1-D/3-D
Imaginary
network!
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Comparison 1-D/3-D
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Influence of proximity to cavity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Real mine
network!
Source depth → ray path
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Ray path → onset polarity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Moment tensor inversion
Amplitude picking
direct wave scattered wave
Introduction
Waveform
modelling
Moment
tensor
inversion
first maximum amplitude = amplitude of the direct wave
direct wave
scattered wave
Summary
energy diffracted
around cavity
?
first maximum amplitude is not always the amplitude of the direct wave
Amplitude vs. waveform inversion
Amplitude inversion:
• homogeneous model of the medium
Introduction
• Green’s functions calculated using ray theory
• inversion of P-wave amplitudes (20-30 amplitudes)
Waveform
modelling
• frequencies: 250-500 Hz
cannot take into account distortion of rays on focal sphere
Moment
tensor
inversion
misinterpretation of amplitudes: which one is the direct wave?
Waveform inversion:
Summary
• 3-D heterogeneous model of the medium
• Green’s functions calculated using FD code
• inversion of full waveforms (15-20 waveforms)
• frequencies (at the moment): 25-100 Hz
inversion is performed in frequency domain
in principle same inversion algorithm as for amplitude inversion, but run
repeatedly for every frequency band (0.5 Hz steps)
Selected explosions: relocation
expl 1
expl 1
expl 3
Introduction
expl 5
Waveform
modelling
expl 3
Moment
tensor
inversion
expl 5
expl 1
Summary
expl 3
expl 5
Explosions (coords in m):
1) x=8306E y=2312N z=-1238
→ 26 m shift
3) x=8218E y=2192N z=-1352
→ 68 m shift
5) x=8214E y=2168N z=-1356
→ 59 m shift
Examples: good fit
Examples: amplitude misfit
Examples: phase misfit
Inversion results
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
explosion 1
explosion 3
explosion 5
Length of time windows
GF duration [10 ms]
Data duration [10 ms]
explosion 1
ISO percentage
GF duration [10 ms]
ISO percentage
GF duration [10 ms]
ISO percentage
Data duration [10 ms]
explosion 3
Data duration [10 ms]
explosion 5
nearly all solutions have high isotropic percentage
best solutions near diagonal
(length of GF time window = length of data time window)
best solutions P –wave + S-wave onset
Explosion 1
Amplitude inversion
Waveform inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
DC = 7%
CLVD = -14%
ISO = 79%
Summary: seismicity in mines
structural model in mines is very complex
Introduction
large and abrupt changes in velocity at
cavities
model varies in time
Waveform
modelling
Moment
tensor
inversion
Summary
earthquake source is complex (single
forces, non-DC components)
small changes in source position lead to
large changes in ray propagation, rays can
be strongly curved
radiated wave field is complex
(reflected, converted, scattered waves, head
waves)
Summary: waveform inversion
In general:
• complex Green’s functions can be calculated by 3-D FD codes (accurate model needed!);
Introduction
• sensitive to time shifts due to mislocation or due to inaccurate velocity model
• frequency band of inverted waves can be easily controlled => stability analysis
Waveform
modelling
In particular:
Moment
tensor
inversion
• good network configuration => focal sphere nicely covered
• inversion algorithm:
• optimal with same window length for Green’s functions and data
Summary
• optimal with simultaneous inversion of P- and S-wave, but excluding S-wave coda
• yields high isotropic percentage, higher than amplitude inversion, almost
independently of window length
promising, but computationally demanding (especially the computation of Green’s
functions with sufficiently small grid point distances)
will be performed for selected events, not whole database
Thank you
for your
attention!
http://commons.wikimedia.org/wiki/File:Preikestolen_Norge.jpg
Slide 8
Moment tensor inversion at
Pyhasalmi ore mine:
accuracy test using explosions
Daniela Kühn (NORSAR)
V. Vavrycuk (Academy of Sciences of the CR)
AIM 2nd annual meeting, 29-30 Sept 2011, Prague, Czech Republic
Pyhäsalmi ore mine, Finland
• microseismic monitoring:
since January 2003
safety of the underground
personnel
optimisation of mining
process
Introduction
Waveform
modelling
Moment
tensor
inversion
• network:
12 1-C geophones
+ 6 3-C geophones (ISS)
3-D geometry
sampling rate: < 3000 Hz
Summary
owned by Inmet Mining Co.
• events:
1500 events /months
(including blasting)
-2 < Mw < 1.5
Inconsistent polarities of P-wave first onset
Moment tensor inversion:
• homogeneous velocity model (as for locations)
• amplitude inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling
Complexity of velocity model
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling: 2D section
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
• E3D: viscoelastic 3-D FD code (Larsen and Schultz, 1995)
• strong interaction with mining cavities: reflection,
scattering, conversion
620 m
• healing of wavefronts
Waveform modelling
synthetic seismograms
- complex waveforms
Introduction
- strong coda
- complex secondary arrivals
Waveform
modelling
- scattering effects stronger
on amplitudes than travel
times,
since
size
of
heterogeneities
(cavities,
access tunnels) same order or
smaller than wavelengths
Moment
tensor
inversion
Summary
arrival times computed by
Eikonal solver still fit
(wavefronts heal quickly
after passing a cavitiy)
observed seismograms
Comparison 1-D/3-D
Imaginary
network!
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Comparison 1-D/3-D
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Influence of proximity to cavity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Real mine
network!
Source depth → ray path
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Ray path → onset polarity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Moment tensor inversion
Amplitude picking
direct wave scattered wave
Introduction
Waveform
modelling
Moment
tensor
inversion
first maximum amplitude = amplitude of the direct wave
direct wave
scattered wave
Summary
energy diffracted
around cavity
?
first maximum amplitude is not always the amplitude of the direct wave
Amplitude vs. waveform inversion
Amplitude inversion:
• homogeneous model of the medium
Introduction
• Green’s functions calculated using ray theory
• inversion of P-wave amplitudes (20-30 amplitudes)
Waveform
modelling
• frequencies: 250-500 Hz
cannot take into account distortion of rays on focal sphere
Moment
tensor
inversion
misinterpretation of amplitudes: which one is the direct wave?
Waveform inversion:
Summary
• 3-D heterogeneous model of the medium
• Green’s functions calculated using FD code
• inversion of full waveforms (15-20 waveforms)
• frequencies (at the moment): 25-100 Hz
inversion is performed in frequency domain
in principle same inversion algorithm as for amplitude inversion, but run
repeatedly for every frequency band (0.5 Hz steps)
Selected explosions: relocation
expl 1
expl 1
expl 3
Introduction
expl 5
Waveform
modelling
expl 3
Moment
tensor
inversion
expl 5
expl 1
Summary
expl 3
expl 5
Explosions (coords in m):
1) x=8306E y=2312N z=-1238
→ 26 m shift
3) x=8218E y=2192N z=-1352
→ 68 m shift
5) x=8214E y=2168N z=-1356
→ 59 m shift
Examples: good fit
Examples: amplitude misfit
Examples: phase misfit
Inversion results
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
explosion 1
explosion 3
explosion 5
Length of time windows
GF duration [10 ms]
Data duration [10 ms]
explosion 1
ISO percentage
GF duration [10 ms]
ISO percentage
GF duration [10 ms]
ISO percentage
Data duration [10 ms]
explosion 3
Data duration [10 ms]
explosion 5
nearly all solutions have high isotropic percentage
best solutions near diagonal
(length of GF time window = length of data time window)
best solutions P –wave + S-wave onset
Explosion 1
Amplitude inversion
Waveform inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
DC = 7%
CLVD = -14%
ISO = 79%
Summary: seismicity in mines
structural model in mines is very complex
Introduction
large and abrupt changes in velocity at
cavities
model varies in time
Waveform
modelling
Moment
tensor
inversion
Summary
earthquake source is complex (single
forces, non-DC components)
small changes in source position lead to
large changes in ray propagation, rays can
be strongly curved
radiated wave field is complex
(reflected, converted, scattered waves, head
waves)
Summary: waveform inversion
In general:
• complex Green’s functions can be calculated by 3-D FD codes (accurate model needed!);
Introduction
• sensitive to time shifts due to mislocation or due to inaccurate velocity model
• frequency band of inverted waves can be easily controlled => stability analysis
Waveform
modelling
In particular:
Moment
tensor
inversion
• good network configuration => focal sphere nicely covered
• inversion algorithm:
• optimal with same window length for Green’s functions and data
Summary
• optimal with simultaneous inversion of P- and S-wave, but excluding S-wave coda
• yields high isotropic percentage, higher than amplitude inversion, almost
independently of window length
promising, but computationally demanding (especially the computation of Green’s
functions with sufficiently small grid point distances)
will be performed for selected events, not whole database
Thank you
for your
attention!
http://commons.wikimedia.org/wiki/File:Preikestolen_Norge.jpg
Slide 9
Moment tensor inversion at
Pyhasalmi ore mine:
accuracy test using explosions
Daniela Kühn (NORSAR)
V. Vavrycuk (Academy of Sciences of the CR)
AIM 2nd annual meeting, 29-30 Sept 2011, Prague, Czech Republic
Pyhäsalmi ore mine, Finland
• microseismic monitoring:
since January 2003
safety of the underground
personnel
optimisation of mining
process
Introduction
Waveform
modelling
Moment
tensor
inversion
• network:
12 1-C geophones
+ 6 3-C geophones (ISS)
3-D geometry
sampling rate: < 3000 Hz
Summary
owned by Inmet Mining Co.
• events:
1500 events /months
(including blasting)
-2 < Mw < 1.5
Inconsistent polarities of P-wave first onset
Moment tensor inversion:
• homogeneous velocity model (as for locations)
• amplitude inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling
Complexity of velocity model
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling: 2D section
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
• E3D: viscoelastic 3-D FD code (Larsen and Schultz, 1995)
• strong interaction with mining cavities: reflection,
scattering, conversion
620 m
• healing of wavefronts
Waveform modelling
synthetic seismograms
- complex waveforms
Introduction
- strong coda
- complex secondary arrivals
Waveform
modelling
- scattering effects stronger
on amplitudes than travel
times,
since
size
of
heterogeneities
(cavities,
access tunnels) same order or
smaller than wavelengths
Moment
tensor
inversion
Summary
arrival times computed by
Eikonal solver still fit
(wavefronts heal quickly
after passing a cavitiy)
observed seismograms
Comparison 1-D/3-D
Imaginary
network!
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Comparison 1-D/3-D
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Influence of proximity to cavity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Real mine
network!
Source depth → ray path
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Ray path → onset polarity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Moment tensor inversion
Amplitude picking
direct wave scattered wave
Introduction
Waveform
modelling
Moment
tensor
inversion
first maximum amplitude = amplitude of the direct wave
direct wave
scattered wave
Summary
energy diffracted
around cavity
?
first maximum amplitude is not always the amplitude of the direct wave
Amplitude vs. waveform inversion
Amplitude inversion:
• homogeneous model of the medium
Introduction
• Green’s functions calculated using ray theory
• inversion of P-wave amplitudes (20-30 amplitudes)
Waveform
modelling
• frequencies: 250-500 Hz
cannot take into account distortion of rays on focal sphere
Moment
tensor
inversion
misinterpretation of amplitudes: which one is the direct wave?
Waveform inversion:
Summary
• 3-D heterogeneous model of the medium
• Green’s functions calculated using FD code
• inversion of full waveforms (15-20 waveforms)
• frequencies (at the moment): 25-100 Hz
inversion is performed in frequency domain
in principle same inversion algorithm as for amplitude inversion, but run
repeatedly for every frequency band (0.5 Hz steps)
Selected explosions: relocation
expl 1
expl 1
expl 3
Introduction
expl 5
Waveform
modelling
expl 3
Moment
tensor
inversion
expl 5
expl 1
Summary
expl 3
expl 5
Explosions (coords in m):
1) x=8306E y=2312N z=-1238
→ 26 m shift
3) x=8218E y=2192N z=-1352
→ 68 m shift
5) x=8214E y=2168N z=-1356
→ 59 m shift
Examples: good fit
Examples: amplitude misfit
Examples: phase misfit
Inversion results
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
explosion 1
explosion 3
explosion 5
Length of time windows
GF duration [10 ms]
Data duration [10 ms]
explosion 1
ISO percentage
GF duration [10 ms]
ISO percentage
GF duration [10 ms]
ISO percentage
Data duration [10 ms]
explosion 3
Data duration [10 ms]
explosion 5
nearly all solutions have high isotropic percentage
best solutions near diagonal
(length of GF time window = length of data time window)
best solutions P –wave + S-wave onset
Explosion 1
Amplitude inversion
Waveform inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
DC = 7%
CLVD = -14%
ISO = 79%
Summary: seismicity in mines
structural model in mines is very complex
Introduction
large and abrupt changes in velocity at
cavities
model varies in time
Waveform
modelling
Moment
tensor
inversion
Summary
earthquake source is complex (single
forces, non-DC components)
small changes in source position lead to
large changes in ray propagation, rays can
be strongly curved
radiated wave field is complex
(reflected, converted, scattered waves, head
waves)
Summary: waveform inversion
In general:
• complex Green’s functions can be calculated by 3-D FD codes (accurate model needed!);
Introduction
• sensitive to time shifts due to mislocation or due to inaccurate velocity model
• frequency band of inverted waves can be easily controlled => stability analysis
Waveform
modelling
In particular:
Moment
tensor
inversion
• good network configuration => focal sphere nicely covered
• inversion algorithm:
• optimal with same window length for Green’s functions and data
Summary
• optimal with simultaneous inversion of P- and S-wave, but excluding S-wave coda
• yields high isotropic percentage, higher than amplitude inversion, almost
independently of window length
promising, but computationally demanding (especially the computation of Green’s
functions with sufficiently small grid point distances)
will be performed for selected events, not whole database
Thank you
for your
attention!
http://commons.wikimedia.org/wiki/File:Preikestolen_Norge.jpg
Slide 10
Moment tensor inversion at
Pyhasalmi ore mine:
accuracy test using explosions
Daniela Kühn (NORSAR)
V. Vavrycuk (Academy of Sciences of the CR)
AIM 2nd annual meeting, 29-30 Sept 2011, Prague, Czech Republic
Pyhäsalmi ore mine, Finland
• microseismic monitoring:
since January 2003
safety of the underground
personnel
optimisation of mining
process
Introduction
Waveform
modelling
Moment
tensor
inversion
• network:
12 1-C geophones
+ 6 3-C geophones (ISS)
3-D geometry
sampling rate: < 3000 Hz
Summary
owned by Inmet Mining Co.
• events:
1500 events /months
(including blasting)
-2 < Mw < 1.5
Inconsistent polarities of P-wave first onset
Moment tensor inversion:
• homogeneous velocity model (as for locations)
• amplitude inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling
Complexity of velocity model
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling: 2D section
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
• E3D: viscoelastic 3-D FD code (Larsen and Schultz, 1995)
• strong interaction with mining cavities: reflection,
scattering, conversion
620 m
• healing of wavefronts
Waveform modelling
synthetic seismograms
- complex waveforms
Introduction
- strong coda
- complex secondary arrivals
Waveform
modelling
- scattering effects stronger
on amplitudes than travel
times,
since
size
of
heterogeneities
(cavities,
access tunnels) same order or
smaller than wavelengths
Moment
tensor
inversion
Summary
arrival times computed by
Eikonal solver still fit
(wavefronts heal quickly
after passing a cavitiy)
observed seismograms
Comparison 1-D/3-D
Imaginary
network!
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Comparison 1-D/3-D
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Influence of proximity to cavity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Real mine
network!
Source depth → ray path
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Ray path → onset polarity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Moment tensor inversion
Amplitude picking
direct wave scattered wave
Introduction
Waveform
modelling
Moment
tensor
inversion
first maximum amplitude = amplitude of the direct wave
direct wave
scattered wave
Summary
energy diffracted
around cavity
?
first maximum amplitude is not always the amplitude of the direct wave
Amplitude vs. waveform inversion
Amplitude inversion:
• homogeneous model of the medium
Introduction
• Green’s functions calculated using ray theory
• inversion of P-wave amplitudes (20-30 amplitudes)
Waveform
modelling
• frequencies: 250-500 Hz
cannot take into account distortion of rays on focal sphere
Moment
tensor
inversion
misinterpretation of amplitudes: which one is the direct wave?
Waveform inversion:
Summary
• 3-D heterogeneous model of the medium
• Green’s functions calculated using FD code
• inversion of full waveforms (15-20 waveforms)
• frequencies (at the moment): 25-100 Hz
inversion is performed in frequency domain
in principle same inversion algorithm as for amplitude inversion, but run
repeatedly for every frequency band (0.5 Hz steps)
Selected explosions: relocation
expl 1
expl 1
expl 3
Introduction
expl 5
Waveform
modelling
expl 3
Moment
tensor
inversion
expl 5
expl 1
Summary
expl 3
expl 5
Explosions (coords in m):
1) x=8306E y=2312N z=-1238
→ 26 m shift
3) x=8218E y=2192N z=-1352
→ 68 m shift
5) x=8214E y=2168N z=-1356
→ 59 m shift
Examples: good fit
Examples: amplitude misfit
Examples: phase misfit
Inversion results
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
explosion 1
explosion 3
explosion 5
Length of time windows
GF duration [10 ms]
Data duration [10 ms]
explosion 1
ISO percentage
GF duration [10 ms]
ISO percentage
GF duration [10 ms]
ISO percentage
Data duration [10 ms]
explosion 3
Data duration [10 ms]
explosion 5
nearly all solutions have high isotropic percentage
best solutions near diagonal
(length of GF time window = length of data time window)
best solutions P –wave + S-wave onset
Explosion 1
Amplitude inversion
Waveform inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
DC = 7%
CLVD = -14%
ISO = 79%
Summary: seismicity in mines
structural model in mines is very complex
Introduction
large and abrupt changes in velocity at
cavities
model varies in time
Waveform
modelling
Moment
tensor
inversion
Summary
earthquake source is complex (single
forces, non-DC components)
small changes in source position lead to
large changes in ray propagation, rays can
be strongly curved
radiated wave field is complex
(reflected, converted, scattered waves, head
waves)
Summary: waveform inversion
In general:
• complex Green’s functions can be calculated by 3-D FD codes (accurate model needed!);
Introduction
• sensitive to time shifts due to mislocation or due to inaccurate velocity model
• frequency band of inverted waves can be easily controlled => stability analysis
Waveform
modelling
In particular:
Moment
tensor
inversion
• good network configuration => focal sphere nicely covered
• inversion algorithm:
• optimal with same window length for Green’s functions and data
Summary
• optimal with simultaneous inversion of P- and S-wave, but excluding S-wave coda
• yields high isotropic percentage, higher than amplitude inversion, almost
independently of window length
promising, but computationally demanding (especially the computation of Green’s
functions with sufficiently small grid point distances)
will be performed for selected events, not whole database
Thank you
for your
attention!
http://commons.wikimedia.org/wiki/File:Preikestolen_Norge.jpg
Slide 11
Moment tensor inversion at
Pyhasalmi ore mine:
accuracy test using explosions
Daniela Kühn (NORSAR)
V. Vavrycuk (Academy of Sciences of the CR)
AIM 2nd annual meeting, 29-30 Sept 2011, Prague, Czech Republic
Pyhäsalmi ore mine, Finland
• microseismic monitoring:
since January 2003
safety of the underground
personnel
optimisation of mining
process
Introduction
Waveform
modelling
Moment
tensor
inversion
• network:
12 1-C geophones
+ 6 3-C geophones (ISS)
3-D geometry
sampling rate: < 3000 Hz
Summary
owned by Inmet Mining Co.
• events:
1500 events /months
(including blasting)
-2 < Mw < 1.5
Inconsistent polarities of P-wave first onset
Moment tensor inversion:
• homogeneous velocity model (as for locations)
• amplitude inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling
Complexity of velocity model
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling: 2D section
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
• E3D: viscoelastic 3-D FD code (Larsen and Schultz, 1995)
• strong interaction with mining cavities: reflection,
scattering, conversion
620 m
• healing of wavefronts
Waveform modelling
synthetic seismograms
- complex waveforms
Introduction
- strong coda
- complex secondary arrivals
Waveform
modelling
- scattering effects stronger
on amplitudes than travel
times,
since
size
of
heterogeneities
(cavities,
access tunnels) same order or
smaller than wavelengths
Moment
tensor
inversion
Summary
arrival times computed by
Eikonal solver still fit
(wavefronts heal quickly
after passing a cavitiy)
observed seismograms
Comparison 1-D/3-D
Imaginary
network!
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Comparison 1-D/3-D
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Influence of proximity to cavity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Real mine
network!
Source depth → ray path
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Ray path → onset polarity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Moment tensor inversion
Amplitude picking
direct wave scattered wave
Introduction
Waveform
modelling
Moment
tensor
inversion
first maximum amplitude = amplitude of the direct wave
direct wave
scattered wave
Summary
energy diffracted
around cavity
?
first maximum amplitude is not always the amplitude of the direct wave
Amplitude vs. waveform inversion
Amplitude inversion:
• homogeneous model of the medium
Introduction
• Green’s functions calculated using ray theory
• inversion of P-wave amplitudes (20-30 amplitudes)
Waveform
modelling
• frequencies: 250-500 Hz
cannot take into account distortion of rays on focal sphere
Moment
tensor
inversion
misinterpretation of amplitudes: which one is the direct wave?
Waveform inversion:
Summary
• 3-D heterogeneous model of the medium
• Green’s functions calculated using FD code
• inversion of full waveforms (15-20 waveforms)
• frequencies (at the moment): 25-100 Hz
inversion is performed in frequency domain
in principle same inversion algorithm as for amplitude inversion, but run
repeatedly for every frequency band (0.5 Hz steps)
Selected explosions: relocation
expl 1
expl 1
expl 3
Introduction
expl 5
Waveform
modelling
expl 3
Moment
tensor
inversion
expl 5
expl 1
Summary
expl 3
expl 5
Explosions (coords in m):
1) x=8306E y=2312N z=-1238
→ 26 m shift
3) x=8218E y=2192N z=-1352
→ 68 m shift
5) x=8214E y=2168N z=-1356
→ 59 m shift
Examples: good fit
Examples: amplitude misfit
Examples: phase misfit
Inversion results
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
explosion 1
explosion 3
explosion 5
Length of time windows
GF duration [10 ms]
Data duration [10 ms]
explosion 1
ISO percentage
GF duration [10 ms]
ISO percentage
GF duration [10 ms]
ISO percentage
Data duration [10 ms]
explosion 3
Data duration [10 ms]
explosion 5
nearly all solutions have high isotropic percentage
best solutions near diagonal
(length of GF time window = length of data time window)
best solutions P –wave + S-wave onset
Explosion 1
Amplitude inversion
Waveform inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
DC = 7%
CLVD = -14%
ISO = 79%
Summary: seismicity in mines
structural model in mines is very complex
Introduction
large and abrupt changes in velocity at
cavities
model varies in time
Waveform
modelling
Moment
tensor
inversion
Summary
earthquake source is complex (single
forces, non-DC components)
small changes in source position lead to
large changes in ray propagation, rays can
be strongly curved
radiated wave field is complex
(reflected, converted, scattered waves, head
waves)
Summary: waveform inversion
In general:
• complex Green’s functions can be calculated by 3-D FD codes (accurate model needed!);
Introduction
• sensitive to time shifts due to mislocation or due to inaccurate velocity model
• frequency band of inverted waves can be easily controlled => stability analysis
Waveform
modelling
In particular:
Moment
tensor
inversion
• good network configuration => focal sphere nicely covered
• inversion algorithm:
• optimal with same window length for Green’s functions and data
Summary
• optimal with simultaneous inversion of P- and S-wave, but excluding S-wave coda
• yields high isotropic percentage, higher than amplitude inversion, almost
independently of window length
promising, but computationally demanding (especially the computation of Green’s
functions with sufficiently small grid point distances)
will be performed for selected events, not whole database
Thank you
for your
attention!
http://commons.wikimedia.org/wiki/File:Preikestolen_Norge.jpg
Slide 12
Moment tensor inversion at
Pyhasalmi ore mine:
accuracy test using explosions
Daniela Kühn (NORSAR)
V. Vavrycuk (Academy of Sciences of the CR)
AIM 2nd annual meeting, 29-30 Sept 2011, Prague, Czech Republic
Pyhäsalmi ore mine, Finland
• microseismic monitoring:
since January 2003
safety of the underground
personnel
optimisation of mining
process
Introduction
Waveform
modelling
Moment
tensor
inversion
• network:
12 1-C geophones
+ 6 3-C geophones (ISS)
3-D geometry
sampling rate: < 3000 Hz
Summary
owned by Inmet Mining Co.
• events:
1500 events /months
(including blasting)
-2 < Mw < 1.5
Inconsistent polarities of P-wave first onset
Moment tensor inversion:
• homogeneous velocity model (as for locations)
• amplitude inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling
Complexity of velocity model
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling: 2D section
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
• E3D: viscoelastic 3-D FD code (Larsen and Schultz, 1995)
• strong interaction with mining cavities: reflection,
scattering, conversion
620 m
• healing of wavefronts
Waveform modelling
synthetic seismograms
- complex waveforms
Introduction
- strong coda
- complex secondary arrivals
Waveform
modelling
- scattering effects stronger
on amplitudes than travel
times,
since
size
of
heterogeneities
(cavities,
access tunnels) same order or
smaller than wavelengths
Moment
tensor
inversion
Summary
arrival times computed by
Eikonal solver still fit
(wavefronts heal quickly
after passing a cavitiy)
observed seismograms
Comparison 1-D/3-D
Imaginary
network!
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Comparison 1-D/3-D
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Influence of proximity to cavity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Real mine
network!
Source depth → ray path
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Ray path → onset polarity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Moment tensor inversion
Amplitude picking
direct wave scattered wave
Introduction
Waveform
modelling
Moment
tensor
inversion
first maximum amplitude = amplitude of the direct wave
direct wave
scattered wave
Summary
energy diffracted
around cavity
?
first maximum amplitude is not always the amplitude of the direct wave
Amplitude vs. waveform inversion
Amplitude inversion:
• homogeneous model of the medium
Introduction
• Green’s functions calculated using ray theory
• inversion of P-wave amplitudes (20-30 amplitudes)
Waveform
modelling
• frequencies: 250-500 Hz
cannot take into account distortion of rays on focal sphere
Moment
tensor
inversion
misinterpretation of amplitudes: which one is the direct wave?
Waveform inversion:
Summary
• 3-D heterogeneous model of the medium
• Green’s functions calculated using FD code
• inversion of full waveforms (15-20 waveforms)
• frequencies (at the moment): 25-100 Hz
inversion is performed in frequency domain
in principle same inversion algorithm as for amplitude inversion, but run
repeatedly for every frequency band (0.5 Hz steps)
Selected explosions: relocation
expl 1
expl 1
expl 3
Introduction
expl 5
Waveform
modelling
expl 3
Moment
tensor
inversion
expl 5
expl 1
Summary
expl 3
expl 5
Explosions (coords in m):
1) x=8306E y=2312N z=-1238
→ 26 m shift
3) x=8218E y=2192N z=-1352
→ 68 m shift
5) x=8214E y=2168N z=-1356
→ 59 m shift
Examples: good fit
Examples: amplitude misfit
Examples: phase misfit
Inversion results
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
explosion 1
explosion 3
explosion 5
Length of time windows
GF duration [10 ms]
Data duration [10 ms]
explosion 1
ISO percentage
GF duration [10 ms]
ISO percentage
GF duration [10 ms]
ISO percentage
Data duration [10 ms]
explosion 3
Data duration [10 ms]
explosion 5
nearly all solutions have high isotropic percentage
best solutions near diagonal
(length of GF time window = length of data time window)
best solutions P –wave + S-wave onset
Explosion 1
Amplitude inversion
Waveform inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
DC = 7%
CLVD = -14%
ISO = 79%
Summary: seismicity in mines
structural model in mines is very complex
Introduction
large and abrupt changes in velocity at
cavities
model varies in time
Waveform
modelling
Moment
tensor
inversion
Summary
earthquake source is complex (single
forces, non-DC components)
small changes in source position lead to
large changes in ray propagation, rays can
be strongly curved
radiated wave field is complex
(reflected, converted, scattered waves, head
waves)
Summary: waveform inversion
In general:
• complex Green’s functions can be calculated by 3-D FD codes (accurate model needed!);
Introduction
• sensitive to time shifts due to mislocation or due to inaccurate velocity model
• frequency band of inverted waves can be easily controlled => stability analysis
Waveform
modelling
In particular:
Moment
tensor
inversion
• good network configuration => focal sphere nicely covered
• inversion algorithm:
• optimal with same window length for Green’s functions and data
Summary
• optimal with simultaneous inversion of P- and S-wave, but excluding S-wave coda
• yields high isotropic percentage, higher than amplitude inversion, almost
independently of window length
promising, but computationally demanding (especially the computation of Green’s
functions with sufficiently small grid point distances)
will be performed for selected events, not whole database
Thank you
for your
attention!
http://commons.wikimedia.org/wiki/File:Preikestolen_Norge.jpg
Slide 13
Moment tensor inversion at
Pyhasalmi ore mine:
accuracy test using explosions
Daniela Kühn (NORSAR)
V. Vavrycuk (Academy of Sciences of the CR)
AIM 2nd annual meeting, 29-30 Sept 2011, Prague, Czech Republic
Pyhäsalmi ore mine, Finland
• microseismic monitoring:
since January 2003
safety of the underground
personnel
optimisation of mining
process
Introduction
Waveform
modelling
Moment
tensor
inversion
• network:
12 1-C geophones
+ 6 3-C geophones (ISS)
3-D geometry
sampling rate: < 3000 Hz
Summary
owned by Inmet Mining Co.
• events:
1500 events /months
(including blasting)
-2 < Mw < 1.5
Inconsistent polarities of P-wave first onset
Moment tensor inversion:
• homogeneous velocity model (as for locations)
• amplitude inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling
Complexity of velocity model
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling: 2D section
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
• E3D: viscoelastic 3-D FD code (Larsen and Schultz, 1995)
• strong interaction with mining cavities: reflection,
scattering, conversion
620 m
• healing of wavefronts
Waveform modelling
synthetic seismograms
- complex waveforms
Introduction
- strong coda
- complex secondary arrivals
Waveform
modelling
- scattering effects stronger
on amplitudes than travel
times,
since
size
of
heterogeneities
(cavities,
access tunnels) same order or
smaller than wavelengths
Moment
tensor
inversion
Summary
arrival times computed by
Eikonal solver still fit
(wavefronts heal quickly
after passing a cavitiy)
observed seismograms
Comparison 1-D/3-D
Imaginary
network!
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Comparison 1-D/3-D
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Influence of proximity to cavity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Real mine
network!
Source depth → ray path
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Ray path → onset polarity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Moment tensor inversion
Amplitude picking
direct wave scattered wave
Introduction
Waveform
modelling
Moment
tensor
inversion
first maximum amplitude = amplitude of the direct wave
direct wave
scattered wave
Summary
energy diffracted
around cavity
?
first maximum amplitude is not always the amplitude of the direct wave
Amplitude vs. waveform inversion
Amplitude inversion:
• homogeneous model of the medium
Introduction
• Green’s functions calculated using ray theory
• inversion of P-wave amplitudes (20-30 amplitudes)
Waveform
modelling
• frequencies: 250-500 Hz
cannot take into account distortion of rays on focal sphere
Moment
tensor
inversion
misinterpretation of amplitudes: which one is the direct wave?
Waveform inversion:
Summary
• 3-D heterogeneous model of the medium
• Green’s functions calculated using FD code
• inversion of full waveforms (15-20 waveforms)
• frequencies (at the moment): 25-100 Hz
inversion is performed in frequency domain
in principle same inversion algorithm as for amplitude inversion, but run
repeatedly for every frequency band (0.5 Hz steps)
Selected explosions: relocation
expl 1
expl 1
expl 3
Introduction
expl 5
Waveform
modelling
expl 3
Moment
tensor
inversion
expl 5
expl 1
Summary
expl 3
expl 5
Explosions (coords in m):
1) x=8306E y=2312N z=-1238
→ 26 m shift
3) x=8218E y=2192N z=-1352
→ 68 m shift
5) x=8214E y=2168N z=-1356
→ 59 m shift
Examples: good fit
Examples: amplitude misfit
Examples: phase misfit
Inversion results
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
explosion 1
explosion 3
explosion 5
Length of time windows
GF duration [10 ms]
Data duration [10 ms]
explosion 1
ISO percentage
GF duration [10 ms]
ISO percentage
GF duration [10 ms]
ISO percentage
Data duration [10 ms]
explosion 3
Data duration [10 ms]
explosion 5
nearly all solutions have high isotropic percentage
best solutions near diagonal
(length of GF time window = length of data time window)
best solutions P –wave + S-wave onset
Explosion 1
Amplitude inversion
Waveform inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
DC = 7%
CLVD = -14%
ISO = 79%
Summary: seismicity in mines
structural model in mines is very complex
Introduction
large and abrupt changes in velocity at
cavities
model varies in time
Waveform
modelling
Moment
tensor
inversion
Summary
earthquake source is complex (single
forces, non-DC components)
small changes in source position lead to
large changes in ray propagation, rays can
be strongly curved
radiated wave field is complex
(reflected, converted, scattered waves, head
waves)
Summary: waveform inversion
In general:
• complex Green’s functions can be calculated by 3-D FD codes (accurate model needed!);
Introduction
• sensitive to time shifts due to mislocation or due to inaccurate velocity model
• frequency band of inverted waves can be easily controlled => stability analysis
Waveform
modelling
In particular:
Moment
tensor
inversion
• good network configuration => focal sphere nicely covered
• inversion algorithm:
• optimal with same window length for Green’s functions and data
Summary
• optimal with simultaneous inversion of P- and S-wave, but excluding S-wave coda
• yields high isotropic percentage, higher than amplitude inversion, almost
independently of window length
promising, but computationally demanding (especially the computation of Green’s
functions with sufficiently small grid point distances)
will be performed for selected events, not whole database
Thank you
for your
attention!
http://commons.wikimedia.org/wiki/File:Preikestolen_Norge.jpg
Slide 14
Moment tensor inversion at
Pyhasalmi ore mine:
accuracy test using explosions
Daniela Kühn (NORSAR)
V. Vavrycuk (Academy of Sciences of the CR)
AIM 2nd annual meeting, 29-30 Sept 2011, Prague, Czech Republic
Pyhäsalmi ore mine, Finland
• microseismic monitoring:
since January 2003
safety of the underground
personnel
optimisation of mining
process
Introduction
Waveform
modelling
Moment
tensor
inversion
• network:
12 1-C geophones
+ 6 3-C geophones (ISS)
3-D geometry
sampling rate: < 3000 Hz
Summary
owned by Inmet Mining Co.
• events:
1500 events /months
(including blasting)
-2 < Mw < 1.5
Inconsistent polarities of P-wave first onset
Moment tensor inversion:
• homogeneous velocity model (as for locations)
• amplitude inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling
Complexity of velocity model
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling: 2D section
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
• E3D: viscoelastic 3-D FD code (Larsen and Schultz, 1995)
• strong interaction with mining cavities: reflection,
scattering, conversion
620 m
• healing of wavefronts
Waveform modelling
synthetic seismograms
- complex waveforms
Introduction
- strong coda
- complex secondary arrivals
Waveform
modelling
- scattering effects stronger
on amplitudes than travel
times,
since
size
of
heterogeneities
(cavities,
access tunnels) same order or
smaller than wavelengths
Moment
tensor
inversion
Summary
arrival times computed by
Eikonal solver still fit
(wavefronts heal quickly
after passing a cavitiy)
observed seismograms
Comparison 1-D/3-D
Imaginary
network!
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Comparison 1-D/3-D
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Influence of proximity to cavity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Real mine
network!
Source depth → ray path
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Ray path → onset polarity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Moment tensor inversion
Amplitude picking
direct wave scattered wave
Introduction
Waveform
modelling
Moment
tensor
inversion
first maximum amplitude = amplitude of the direct wave
direct wave
scattered wave
Summary
energy diffracted
around cavity
?
first maximum amplitude is not always the amplitude of the direct wave
Amplitude vs. waveform inversion
Amplitude inversion:
• homogeneous model of the medium
Introduction
• Green’s functions calculated using ray theory
• inversion of P-wave amplitudes (20-30 amplitudes)
Waveform
modelling
• frequencies: 250-500 Hz
cannot take into account distortion of rays on focal sphere
Moment
tensor
inversion
misinterpretation of amplitudes: which one is the direct wave?
Waveform inversion:
Summary
• 3-D heterogeneous model of the medium
• Green’s functions calculated using FD code
• inversion of full waveforms (15-20 waveforms)
• frequencies (at the moment): 25-100 Hz
inversion is performed in frequency domain
in principle same inversion algorithm as for amplitude inversion, but run
repeatedly for every frequency band (0.5 Hz steps)
Selected explosions: relocation
expl 1
expl 1
expl 3
Introduction
expl 5
Waveform
modelling
expl 3
Moment
tensor
inversion
expl 5
expl 1
Summary
expl 3
expl 5
Explosions (coords in m):
1) x=8306E y=2312N z=-1238
→ 26 m shift
3) x=8218E y=2192N z=-1352
→ 68 m shift
5) x=8214E y=2168N z=-1356
→ 59 m shift
Examples: good fit
Examples: amplitude misfit
Examples: phase misfit
Inversion results
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
explosion 1
explosion 3
explosion 5
Length of time windows
GF duration [10 ms]
Data duration [10 ms]
explosion 1
ISO percentage
GF duration [10 ms]
ISO percentage
GF duration [10 ms]
ISO percentage
Data duration [10 ms]
explosion 3
Data duration [10 ms]
explosion 5
nearly all solutions have high isotropic percentage
best solutions near diagonal
(length of GF time window = length of data time window)
best solutions P –wave + S-wave onset
Explosion 1
Amplitude inversion
Waveform inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
DC = 7%
CLVD = -14%
ISO = 79%
Summary: seismicity in mines
structural model in mines is very complex
Introduction
large and abrupt changes in velocity at
cavities
model varies in time
Waveform
modelling
Moment
tensor
inversion
Summary
earthquake source is complex (single
forces, non-DC components)
small changes in source position lead to
large changes in ray propagation, rays can
be strongly curved
radiated wave field is complex
(reflected, converted, scattered waves, head
waves)
Summary: waveform inversion
In general:
• complex Green’s functions can be calculated by 3-D FD codes (accurate model needed!);
Introduction
• sensitive to time shifts due to mislocation or due to inaccurate velocity model
• frequency band of inverted waves can be easily controlled => stability analysis
Waveform
modelling
In particular:
Moment
tensor
inversion
• good network configuration => focal sphere nicely covered
• inversion algorithm:
• optimal with same window length for Green’s functions and data
Summary
• optimal with simultaneous inversion of P- and S-wave, but excluding S-wave coda
• yields high isotropic percentage, higher than amplitude inversion, almost
independently of window length
promising, but computationally demanding (especially the computation of Green’s
functions with sufficiently small grid point distances)
will be performed for selected events, not whole database
Thank you
for your
attention!
http://commons.wikimedia.org/wiki/File:Preikestolen_Norge.jpg
Slide 15
Moment tensor inversion at
Pyhasalmi ore mine:
accuracy test using explosions
Daniela Kühn (NORSAR)
V. Vavrycuk (Academy of Sciences of the CR)
AIM 2nd annual meeting, 29-30 Sept 2011, Prague, Czech Republic
Pyhäsalmi ore mine, Finland
• microseismic monitoring:
since January 2003
safety of the underground
personnel
optimisation of mining
process
Introduction
Waveform
modelling
Moment
tensor
inversion
• network:
12 1-C geophones
+ 6 3-C geophones (ISS)
3-D geometry
sampling rate: < 3000 Hz
Summary
owned by Inmet Mining Co.
• events:
1500 events /months
(including blasting)
-2 < Mw < 1.5
Inconsistent polarities of P-wave first onset
Moment tensor inversion:
• homogeneous velocity model (as for locations)
• amplitude inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling
Complexity of velocity model
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling: 2D section
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
• E3D: viscoelastic 3-D FD code (Larsen and Schultz, 1995)
• strong interaction with mining cavities: reflection,
scattering, conversion
620 m
• healing of wavefronts
Waveform modelling
synthetic seismograms
- complex waveforms
Introduction
- strong coda
- complex secondary arrivals
Waveform
modelling
- scattering effects stronger
on amplitudes than travel
times,
since
size
of
heterogeneities
(cavities,
access tunnels) same order or
smaller than wavelengths
Moment
tensor
inversion
Summary
arrival times computed by
Eikonal solver still fit
(wavefronts heal quickly
after passing a cavitiy)
observed seismograms
Comparison 1-D/3-D
Imaginary
network!
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Comparison 1-D/3-D
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Influence of proximity to cavity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Real mine
network!
Source depth → ray path
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Ray path → onset polarity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Moment tensor inversion
Amplitude picking
direct wave scattered wave
Introduction
Waveform
modelling
Moment
tensor
inversion
first maximum amplitude = amplitude of the direct wave
direct wave
scattered wave
Summary
energy diffracted
around cavity
?
first maximum amplitude is not always the amplitude of the direct wave
Amplitude vs. waveform inversion
Amplitude inversion:
• homogeneous model of the medium
Introduction
• Green’s functions calculated using ray theory
• inversion of P-wave amplitudes (20-30 amplitudes)
Waveform
modelling
• frequencies: 250-500 Hz
cannot take into account distortion of rays on focal sphere
Moment
tensor
inversion
misinterpretation of amplitudes: which one is the direct wave?
Waveform inversion:
Summary
• 3-D heterogeneous model of the medium
• Green’s functions calculated using FD code
• inversion of full waveforms (15-20 waveforms)
• frequencies (at the moment): 25-100 Hz
inversion is performed in frequency domain
in principle same inversion algorithm as for amplitude inversion, but run
repeatedly for every frequency band (0.5 Hz steps)
Selected explosions: relocation
expl 1
expl 1
expl 3
Introduction
expl 5
Waveform
modelling
expl 3
Moment
tensor
inversion
expl 5
expl 1
Summary
expl 3
expl 5
Explosions (coords in m):
1) x=8306E y=2312N z=-1238
→ 26 m shift
3) x=8218E y=2192N z=-1352
→ 68 m shift
5) x=8214E y=2168N z=-1356
→ 59 m shift
Examples: good fit
Examples: amplitude misfit
Examples: phase misfit
Inversion results
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
explosion 1
explosion 3
explosion 5
Length of time windows
GF duration [10 ms]
Data duration [10 ms]
explosion 1
ISO percentage
GF duration [10 ms]
ISO percentage
GF duration [10 ms]
ISO percentage
Data duration [10 ms]
explosion 3
Data duration [10 ms]
explosion 5
nearly all solutions have high isotropic percentage
best solutions near diagonal
(length of GF time window = length of data time window)
best solutions P –wave + S-wave onset
Explosion 1
Amplitude inversion
Waveform inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
DC = 7%
CLVD = -14%
ISO = 79%
Summary: seismicity in mines
structural model in mines is very complex
Introduction
large and abrupt changes in velocity at
cavities
model varies in time
Waveform
modelling
Moment
tensor
inversion
Summary
earthquake source is complex (single
forces, non-DC components)
small changes in source position lead to
large changes in ray propagation, rays can
be strongly curved
radiated wave field is complex
(reflected, converted, scattered waves, head
waves)
Summary: waveform inversion
In general:
• complex Green’s functions can be calculated by 3-D FD codes (accurate model needed!);
Introduction
• sensitive to time shifts due to mislocation or due to inaccurate velocity model
• frequency band of inverted waves can be easily controlled => stability analysis
Waveform
modelling
In particular:
Moment
tensor
inversion
• good network configuration => focal sphere nicely covered
• inversion algorithm:
• optimal with same window length for Green’s functions and data
Summary
• optimal with simultaneous inversion of P- and S-wave, but excluding S-wave coda
• yields high isotropic percentage, higher than amplitude inversion, almost
independently of window length
promising, but computationally demanding (especially the computation of Green’s
functions with sufficiently small grid point distances)
will be performed for selected events, not whole database
Thank you
for your
attention!
http://commons.wikimedia.org/wiki/File:Preikestolen_Norge.jpg
Slide 16
Moment tensor inversion at
Pyhasalmi ore mine:
accuracy test using explosions
Daniela Kühn (NORSAR)
V. Vavrycuk (Academy of Sciences of the CR)
AIM 2nd annual meeting, 29-30 Sept 2011, Prague, Czech Republic
Pyhäsalmi ore mine, Finland
• microseismic monitoring:
since January 2003
safety of the underground
personnel
optimisation of mining
process
Introduction
Waveform
modelling
Moment
tensor
inversion
• network:
12 1-C geophones
+ 6 3-C geophones (ISS)
3-D geometry
sampling rate: < 3000 Hz
Summary
owned by Inmet Mining Co.
• events:
1500 events /months
(including blasting)
-2 < Mw < 1.5
Inconsistent polarities of P-wave first onset
Moment tensor inversion:
• homogeneous velocity model (as for locations)
• amplitude inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling
Complexity of velocity model
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling: 2D section
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
• E3D: viscoelastic 3-D FD code (Larsen and Schultz, 1995)
• strong interaction with mining cavities: reflection,
scattering, conversion
620 m
• healing of wavefronts
Waveform modelling
synthetic seismograms
- complex waveforms
Introduction
- strong coda
- complex secondary arrivals
Waveform
modelling
- scattering effects stronger
on amplitudes than travel
times,
since
size
of
heterogeneities
(cavities,
access tunnels) same order or
smaller than wavelengths
Moment
tensor
inversion
Summary
arrival times computed by
Eikonal solver still fit
(wavefronts heal quickly
after passing a cavitiy)
observed seismograms
Comparison 1-D/3-D
Imaginary
network!
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Comparison 1-D/3-D
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Influence of proximity to cavity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Real mine
network!
Source depth → ray path
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Ray path → onset polarity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Moment tensor inversion
Amplitude picking
direct wave scattered wave
Introduction
Waveform
modelling
Moment
tensor
inversion
first maximum amplitude = amplitude of the direct wave
direct wave
scattered wave
Summary
energy diffracted
around cavity
?
first maximum amplitude is not always the amplitude of the direct wave
Amplitude vs. waveform inversion
Amplitude inversion:
• homogeneous model of the medium
Introduction
• Green’s functions calculated using ray theory
• inversion of P-wave amplitudes (20-30 amplitudes)
Waveform
modelling
• frequencies: 250-500 Hz
cannot take into account distortion of rays on focal sphere
Moment
tensor
inversion
misinterpretation of amplitudes: which one is the direct wave?
Waveform inversion:
Summary
• 3-D heterogeneous model of the medium
• Green’s functions calculated using FD code
• inversion of full waveforms (15-20 waveforms)
• frequencies (at the moment): 25-100 Hz
inversion is performed in frequency domain
in principle same inversion algorithm as for amplitude inversion, but run
repeatedly for every frequency band (0.5 Hz steps)
Selected explosions: relocation
expl 1
expl 1
expl 3
Introduction
expl 5
Waveform
modelling
expl 3
Moment
tensor
inversion
expl 5
expl 1
Summary
expl 3
expl 5
Explosions (coords in m):
1) x=8306E y=2312N z=-1238
→ 26 m shift
3) x=8218E y=2192N z=-1352
→ 68 m shift
5) x=8214E y=2168N z=-1356
→ 59 m shift
Examples: good fit
Examples: amplitude misfit
Examples: phase misfit
Inversion results
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
explosion 1
explosion 3
explosion 5
Length of time windows
GF duration [10 ms]
Data duration [10 ms]
explosion 1
ISO percentage
GF duration [10 ms]
ISO percentage
GF duration [10 ms]
ISO percentage
Data duration [10 ms]
explosion 3
Data duration [10 ms]
explosion 5
nearly all solutions have high isotropic percentage
best solutions near diagonal
(length of GF time window = length of data time window)
best solutions P –wave + S-wave onset
Explosion 1
Amplitude inversion
Waveform inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
DC = 7%
CLVD = -14%
ISO = 79%
Summary: seismicity in mines
structural model in mines is very complex
Introduction
large and abrupt changes in velocity at
cavities
model varies in time
Waveform
modelling
Moment
tensor
inversion
Summary
earthquake source is complex (single
forces, non-DC components)
small changes in source position lead to
large changes in ray propagation, rays can
be strongly curved
radiated wave field is complex
(reflected, converted, scattered waves, head
waves)
Summary: waveform inversion
In general:
• complex Green’s functions can be calculated by 3-D FD codes (accurate model needed!);
Introduction
• sensitive to time shifts due to mislocation or due to inaccurate velocity model
• frequency band of inverted waves can be easily controlled => stability analysis
Waveform
modelling
In particular:
Moment
tensor
inversion
• good network configuration => focal sphere nicely covered
• inversion algorithm:
• optimal with same window length for Green’s functions and data
Summary
• optimal with simultaneous inversion of P- and S-wave, but excluding S-wave coda
• yields high isotropic percentage, higher than amplitude inversion, almost
independently of window length
promising, but computationally demanding (especially the computation of Green’s
functions with sufficiently small grid point distances)
will be performed for selected events, not whole database
Thank you
for your
attention!
http://commons.wikimedia.org/wiki/File:Preikestolen_Norge.jpg
Slide 17
Moment tensor inversion at
Pyhasalmi ore mine:
accuracy test using explosions
Daniela Kühn (NORSAR)
V. Vavrycuk (Academy of Sciences of the CR)
AIM 2nd annual meeting, 29-30 Sept 2011, Prague, Czech Republic
Pyhäsalmi ore mine, Finland
• microseismic monitoring:
since January 2003
safety of the underground
personnel
optimisation of mining
process
Introduction
Waveform
modelling
Moment
tensor
inversion
• network:
12 1-C geophones
+ 6 3-C geophones (ISS)
3-D geometry
sampling rate: < 3000 Hz
Summary
owned by Inmet Mining Co.
• events:
1500 events /months
(including blasting)
-2 < Mw < 1.5
Inconsistent polarities of P-wave first onset
Moment tensor inversion:
• homogeneous velocity model (as for locations)
• amplitude inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling
Complexity of velocity model
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling: 2D section
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
• E3D: viscoelastic 3-D FD code (Larsen and Schultz, 1995)
• strong interaction with mining cavities: reflection,
scattering, conversion
620 m
• healing of wavefronts
Waveform modelling
synthetic seismograms
- complex waveforms
Introduction
- strong coda
- complex secondary arrivals
Waveform
modelling
- scattering effects stronger
on amplitudes than travel
times,
since
size
of
heterogeneities
(cavities,
access tunnels) same order or
smaller than wavelengths
Moment
tensor
inversion
Summary
arrival times computed by
Eikonal solver still fit
(wavefronts heal quickly
after passing a cavitiy)
observed seismograms
Comparison 1-D/3-D
Imaginary
network!
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Comparison 1-D/3-D
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Influence of proximity to cavity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Real mine
network!
Source depth → ray path
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Ray path → onset polarity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Moment tensor inversion
Amplitude picking
direct wave scattered wave
Introduction
Waveform
modelling
Moment
tensor
inversion
first maximum amplitude = amplitude of the direct wave
direct wave
scattered wave
Summary
energy diffracted
around cavity
?
first maximum amplitude is not always the amplitude of the direct wave
Amplitude vs. waveform inversion
Amplitude inversion:
• homogeneous model of the medium
Introduction
• Green’s functions calculated using ray theory
• inversion of P-wave amplitudes (20-30 amplitudes)
Waveform
modelling
• frequencies: 250-500 Hz
cannot take into account distortion of rays on focal sphere
Moment
tensor
inversion
misinterpretation of amplitudes: which one is the direct wave?
Waveform inversion:
Summary
• 3-D heterogeneous model of the medium
• Green’s functions calculated using FD code
• inversion of full waveforms (15-20 waveforms)
• frequencies (at the moment): 25-100 Hz
inversion is performed in frequency domain
in principle same inversion algorithm as for amplitude inversion, but run
repeatedly for every frequency band (0.5 Hz steps)
Selected explosions: relocation
expl 1
expl 1
expl 3
Introduction
expl 5
Waveform
modelling
expl 3
Moment
tensor
inversion
expl 5
expl 1
Summary
expl 3
expl 5
Explosions (coords in m):
1) x=8306E y=2312N z=-1238
→ 26 m shift
3) x=8218E y=2192N z=-1352
→ 68 m shift
5) x=8214E y=2168N z=-1356
→ 59 m shift
Examples: good fit
Examples: amplitude misfit
Examples: phase misfit
Inversion results
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
explosion 1
explosion 3
explosion 5
Length of time windows
GF duration [10 ms]
Data duration [10 ms]
explosion 1
ISO percentage
GF duration [10 ms]
ISO percentage
GF duration [10 ms]
ISO percentage
Data duration [10 ms]
explosion 3
Data duration [10 ms]
explosion 5
nearly all solutions have high isotropic percentage
best solutions near diagonal
(length of GF time window = length of data time window)
best solutions P –wave + S-wave onset
Explosion 1
Amplitude inversion
Waveform inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
DC = 7%
CLVD = -14%
ISO = 79%
Summary: seismicity in mines
structural model in mines is very complex
Introduction
large and abrupt changes in velocity at
cavities
model varies in time
Waveform
modelling
Moment
tensor
inversion
Summary
earthquake source is complex (single
forces, non-DC components)
small changes in source position lead to
large changes in ray propagation, rays can
be strongly curved
radiated wave field is complex
(reflected, converted, scattered waves, head
waves)
Summary: waveform inversion
In general:
• complex Green’s functions can be calculated by 3-D FD codes (accurate model needed!);
Introduction
• sensitive to time shifts due to mislocation or due to inaccurate velocity model
• frequency band of inverted waves can be easily controlled => stability analysis
Waveform
modelling
In particular:
Moment
tensor
inversion
• good network configuration => focal sphere nicely covered
• inversion algorithm:
• optimal with same window length for Green’s functions and data
Summary
• optimal with simultaneous inversion of P- and S-wave, but excluding S-wave coda
• yields high isotropic percentage, higher than amplitude inversion, almost
independently of window length
promising, but computationally demanding (especially the computation of Green’s
functions with sufficiently small grid point distances)
will be performed for selected events, not whole database
Thank you
for your
attention!
http://commons.wikimedia.org/wiki/File:Preikestolen_Norge.jpg
Slide 18
Moment tensor inversion at
Pyhasalmi ore mine:
accuracy test using explosions
Daniela Kühn (NORSAR)
V. Vavrycuk (Academy of Sciences of the CR)
AIM 2nd annual meeting, 29-30 Sept 2011, Prague, Czech Republic
Pyhäsalmi ore mine, Finland
• microseismic monitoring:
since January 2003
safety of the underground
personnel
optimisation of mining
process
Introduction
Waveform
modelling
Moment
tensor
inversion
• network:
12 1-C geophones
+ 6 3-C geophones (ISS)
3-D geometry
sampling rate: < 3000 Hz
Summary
owned by Inmet Mining Co.
• events:
1500 events /months
(including blasting)
-2 < Mw < 1.5
Inconsistent polarities of P-wave first onset
Moment tensor inversion:
• homogeneous velocity model (as for locations)
• amplitude inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling
Complexity of velocity model
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling: 2D section
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
• E3D: viscoelastic 3-D FD code (Larsen and Schultz, 1995)
• strong interaction with mining cavities: reflection,
scattering, conversion
620 m
• healing of wavefronts
Waveform modelling
synthetic seismograms
- complex waveforms
Introduction
- strong coda
- complex secondary arrivals
Waveform
modelling
- scattering effects stronger
on amplitudes than travel
times,
since
size
of
heterogeneities
(cavities,
access tunnels) same order or
smaller than wavelengths
Moment
tensor
inversion
Summary
arrival times computed by
Eikonal solver still fit
(wavefronts heal quickly
after passing a cavitiy)
observed seismograms
Comparison 1-D/3-D
Imaginary
network!
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Comparison 1-D/3-D
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Influence of proximity to cavity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Real mine
network!
Source depth → ray path
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Ray path → onset polarity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Moment tensor inversion
Amplitude picking
direct wave scattered wave
Introduction
Waveform
modelling
Moment
tensor
inversion
first maximum amplitude = amplitude of the direct wave
direct wave
scattered wave
Summary
energy diffracted
around cavity
?
first maximum amplitude is not always the amplitude of the direct wave
Amplitude vs. waveform inversion
Amplitude inversion:
• homogeneous model of the medium
Introduction
• Green’s functions calculated using ray theory
• inversion of P-wave amplitudes (20-30 amplitudes)
Waveform
modelling
• frequencies: 250-500 Hz
cannot take into account distortion of rays on focal sphere
Moment
tensor
inversion
misinterpretation of amplitudes: which one is the direct wave?
Waveform inversion:
Summary
• 3-D heterogeneous model of the medium
• Green’s functions calculated using FD code
• inversion of full waveforms (15-20 waveforms)
• frequencies (at the moment): 25-100 Hz
inversion is performed in frequency domain
in principle same inversion algorithm as for amplitude inversion, but run
repeatedly for every frequency band (0.5 Hz steps)
Selected explosions: relocation
expl 1
expl 1
expl 3
Introduction
expl 5
Waveform
modelling
expl 3
Moment
tensor
inversion
expl 5
expl 1
Summary
expl 3
expl 5
Explosions (coords in m):
1) x=8306E y=2312N z=-1238
→ 26 m shift
3) x=8218E y=2192N z=-1352
→ 68 m shift
5) x=8214E y=2168N z=-1356
→ 59 m shift
Examples: good fit
Examples: amplitude misfit
Examples: phase misfit
Inversion results
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
explosion 1
explosion 3
explosion 5
Length of time windows
GF duration [10 ms]
Data duration [10 ms]
explosion 1
ISO percentage
GF duration [10 ms]
ISO percentage
GF duration [10 ms]
ISO percentage
Data duration [10 ms]
explosion 3
Data duration [10 ms]
explosion 5
nearly all solutions have high isotropic percentage
best solutions near diagonal
(length of GF time window = length of data time window)
best solutions P –wave + S-wave onset
Explosion 1
Amplitude inversion
Waveform inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
DC = 7%
CLVD = -14%
ISO = 79%
Summary: seismicity in mines
structural model in mines is very complex
Introduction
large and abrupt changes in velocity at
cavities
model varies in time
Waveform
modelling
Moment
tensor
inversion
Summary
earthquake source is complex (single
forces, non-DC components)
small changes in source position lead to
large changes in ray propagation, rays can
be strongly curved
radiated wave field is complex
(reflected, converted, scattered waves, head
waves)
Summary: waveform inversion
In general:
• complex Green’s functions can be calculated by 3-D FD codes (accurate model needed!);
Introduction
• sensitive to time shifts due to mislocation or due to inaccurate velocity model
• frequency band of inverted waves can be easily controlled => stability analysis
Waveform
modelling
In particular:
Moment
tensor
inversion
• good network configuration => focal sphere nicely covered
• inversion algorithm:
• optimal with same window length for Green’s functions and data
Summary
• optimal with simultaneous inversion of P- and S-wave, but excluding S-wave coda
• yields high isotropic percentage, higher than amplitude inversion, almost
independently of window length
promising, but computationally demanding (especially the computation of Green’s
functions with sufficiently small grid point distances)
will be performed for selected events, not whole database
Thank you
for your
attention!
http://commons.wikimedia.org/wiki/File:Preikestolen_Norge.jpg
Slide 19
Moment tensor inversion at
Pyhasalmi ore mine:
accuracy test using explosions
Daniela Kühn (NORSAR)
V. Vavrycuk (Academy of Sciences of the CR)
AIM 2nd annual meeting, 29-30 Sept 2011, Prague, Czech Republic
Pyhäsalmi ore mine, Finland
• microseismic monitoring:
since January 2003
safety of the underground
personnel
optimisation of mining
process
Introduction
Waveform
modelling
Moment
tensor
inversion
• network:
12 1-C geophones
+ 6 3-C geophones (ISS)
3-D geometry
sampling rate: < 3000 Hz
Summary
owned by Inmet Mining Co.
• events:
1500 events /months
(including blasting)
-2 < Mw < 1.5
Inconsistent polarities of P-wave first onset
Moment tensor inversion:
• homogeneous velocity model (as for locations)
• amplitude inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling
Complexity of velocity model
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling: 2D section
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
• E3D: viscoelastic 3-D FD code (Larsen and Schultz, 1995)
• strong interaction with mining cavities: reflection,
scattering, conversion
620 m
• healing of wavefronts
Waveform modelling
synthetic seismograms
- complex waveforms
Introduction
- strong coda
- complex secondary arrivals
Waveform
modelling
- scattering effects stronger
on amplitudes than travel
times,
since
size
of
heterogeneities
(cavities,
access tunnels) same order or
smaller than wavelengths
Moment
tensor
inversion
Summary
arrival times computed by
Eikonal solver still fit
(wavefronts heal quickly
after passing a cavitiy)
observed seismograms
Comparison 1-D/3-D
Imaginary
network!
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Comparison 1-D/3-D
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Influence of proximity to cavity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Real mine
network!
Source depth → ray path
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Ray path → onset polarity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Moment tensor inversion
Amplitude picking
direct wave scattered wave
Introduction
Waveform
modelling
Moment
tensor
inversion
first maximum amplitude = amplitude of the direct wave
direct wave
scattered wave
Summary
energy diffracted
around cavity
?
first maximum amplitude is not always the amplitude of the direct wave
Amplitude vs. waveform inversion
Amplitude inversion:
• homogeneous model of the medium
Introduction
• Green’s functions calculated using ray theory
• inversion of P-wave amplitudes (20-30 amplitudes)
Waveform
modelling
• frequencies: 250-500 Hz
cannot take into account distortion of rays on focal sphere
Moment
tensor
inversion
misinterpretation of amplitudes: which one is the direct wave?
Waveform inversion:
Summary
• 3-D heterogeneous model of the medium
• Green’s functions calculated using FD code
• inversion of full waveforms (15-20 waveforms)
• frequencies (at the moment): 25-100 Hz
inversion is performed in frequency domain
in principle same inversion algorithm as for amplitude inversion, but run
repeatedly for every frequency band (0.5 Hz steps)
Selected explosions: relocation
expl 1
expl 1
expl 3
Introduction
expl 5
Waveform
modelling
expl 3
Moment
tensor
inversion
expl 5
expl 1
Summary
expl 3
expl 5
Explosions (coords in m):
1) x=8306E y=2312N z=-1238
→ 26 m shift
3) x=8218E y=2192N z=-1352
→ 68 m shift
5) x=8214E y=2168N z=-1356
→ 59 m shift
Examples: good fit
Examples: amplitude misfit
Examples: phase misfit
Inversion results
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
explosion 1
explosion 3
explosion 5
Length of time windows
GF duration [10 ms]
Data duration [10 ms]
explosion 1
ISO percentage
GF duration [10 ms]
ISO percentage
GF duration [10 ms]
ISO percentage
Data duration [10 ms]
explosion 3
Data duration [10 ms]
explosion 5
nearly all solutions have high isotropic percentage
best solutions near diagonal
(length of GF time window = length of data time window)
best solutions P –wave + S-wave onset
Explosion 1
Amplitude inversion
Waveform inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
DC = 7%
CLVD = -14%
ISO = 79%
Summary: seismicity in mines
structural model in mines is very complex
Introduction
large and abrupt changes in velocity at
cavities
model varies in time
Waveform
modelling
Moment
tensor
inversion
Summary
earthquake source is complex (single
forces, non-DC components)
small changes in source position lead to
large changes in ray propagation, rays can
be strongly curved
radiated wave field is complex
(reflected, converted, scattered waves, head
waves)
Summary: waveform inversion
In general:
• complex Green’s functions can be calculated by 3-D FD codes (accurate model needed!);
Introduction
• sensitive to time shifts due to mislocation or due to inaccurate velocity model
• frequency band of inverted waves can be easily controlled => stability analysis
Waveform
modelling
In particular:
Moment
tensor
inversion
• good network configuration => focal sphere nicely covered
• inversion algorithm:
• optimal with same window length for Green’s functions and data
Summary
• optimal with simultaneous inversion of P- and S-wave, but excluding S-wave coda
• yields high isotropic percentage, higher than amplitude inversion, almost
independently of window length
promising, but computationally demanding (especially the computation of Green’s
functions with sufficiently small grid point distances)
will be performed for selected events, not whole database
Thank you
for your
attention!
http://commons.wikimedia.org/wiki/File:Preikestolen_Norge.jpg
Slide 20
Moment tensor inversion at
Pyhasalmi ore mine:
accuracy test using explosions
Daniela Kühn (NORSAR)
V. Vavrycuk (Academy of Sciences of the CR)
AIM 2nd annual meeting, 29-30 Sept 2011, Prague, Czech Republic
Pyhäsalmi ore mine, Finland
• microseismic monitoring:
since January 2003
safety of the underground
personnel
optimisation of mining
process
Introduction
Waveform
modelling
Moment
tensor
inversion
• network:
12 1-C geophones
+ 6 3-C geophones (ISS)
3-D geometry
sampling rate: < 3000 Hz
Summary
owned by Inmet Mining Co.
• events:
1500 events /months
(including blasting)
-2 < Mw < 1.5
Inconsistent polarities of P-wave first onset
Moment tensor inversion:
• homogeneous velocity model (as for locations)
• amplitude inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling
Complexity of velocity model
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling: 2D section
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
• E3D: viscoelastic 3-D FD code (Larsen and Schultz, 1995)
• strong interaction with mining cavities: reflection,
scattering, conversion
620 m
• healing of wavefronts
Waveform modelling
synthetic seismograms
- complex waveforms
Introduction
- strong coda
- complex secondary arrivals
Waveform
modelling
- scattering effects stronger
on amplitudes than travel
times,
since
size
of
heterogeneities
(cavities,
access tunnels) same order or
smaller than wavelengths
Moment
tensor
inversion
Summary
arrival times computed by
Eikonal solver still fit
(wavefronts heal quickly
after passing a cavitiy)
observed seismograms
Comparison 1-D/3-D
Imaginary
network!
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Comparison 1-D/3-D
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Influence of proximity to cavity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Real mine
network!
Source depth → ray path
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Ray path → onset polarity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Moment tensor inversion
Amplitude picking
direct wave scattered wave
Introduction
Waveform
modelling
Moment
tensor
inversion
first maximum amplitude = amplitude of the direct wave
direct wave
scattered wave
Summary
energy diffracted
around cavity
?
first maximum amplitude is not always the amplitude of the direct wave
Amplitude vs. waveform inversion
Amplitude inversion:
• homogeneous model of the medium
Introduction
• Green’s functions calculated using ray theory
• inversion of P-wave amplitudes (20-30 amplitudes)
Waveform
modelling
• frequencies: 250-500 Hz
cannot take into account distortion of rays on focal sphere
Moment
tensor
inversion
misinterpretation of amplitudes: which one is the direct wave?
Waveform inversion:
Summary
• 3-D heterogeneous model of the medium
• Green’s functions calculated using FD code
• inversion of full waveforms (15-20 waveforms)
• frequencies (at the moment): 25-100 Hz
inversion is performed in frequency domain
in principle same inversion algorithm as for amplitude inversion, but run
repeatedly for every frequency band (0.5 Hz steps)
Selected explosions: relocation
expl 1
expl 1
expl 3
Introduction
expl 5
Waveform
modelling
expl 3
Moment
tensor
inversion
expl 5
expl 1
Summary
expl 3
expl 5
Explosions (coords in m):
1) x=8306E y=2312N z=-1238
→ 26 m shift
3) x=8218E y=2192N z=-1352
→ 68 m shift
5) x=8214E y=2168N z=-1356
→ 59 m shift
Examples: good fit
Examples: amplitude misfit
Examples: phase misfit
Inversion results
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
explosion 1
explosion 3
explosion 5
Length of time windows
GF duration [10 ms]
Data duration [10 ms]
explosion 1
ISO percentage
GF duration [10 ms]
ISO percentage
GF duration [10 ms]
ISO percentage
Data duration [10 ms]
explosion 3
Data duration [10 ms]
explosion 5
nearly all solutions have high isotropic percentage
best solutions near diagonal
(length of GF time window = length of data time window)
best solutions P –wave + S-wave onset
Explosion 1
Amplitude inversion
Waveform inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
DC = 7%
CLVD = -14%
ISO = 79%
Summary: seismicity in mines
structural model in mines is very complex
Introduction
large and abrupt changes in velocity at
cavities
model varies in time
Waveform
modelling
Moment
tensor
inversion
Summary
earthquake source is complex (single
forces, non-DC components)
small changes in source position lead to
large changes in ray propagation, rays can
be strongly curved
radiated wave field is complex
(reflected, converted, scattered waves, head
waves)
Summary: waveform inversion
In general:
• complex Green’s functions can be calculated by 3-D FD codes (accurate model needed!);
Introduction
• sensitive to time shifts due to mislocation or due to inaccurate velocity model
• frequency band of inverted waves can be easily controlled => stability analysis
Waveform
modelling
In particular:
Moment
tensor
inversion
• good network configuration => focal sphere nicely covered
• inversion algorithm:
• optimal with same window length for Green’s functions and data
Summary
• optimal with simultaneous inversion of P- and S-wave, but excluding S-wave coda
• yields high isotropic percentage, higher than amplitude inversion, almost
independently of window length
promising, but computationally demanding (especially the computation of Green’s
functions with sufficiently small grid point distances)
will be performed for selected events, not whole database
Thank you
for your
attention!
http://commons.wikimedia.org/wiki/File:Preikestolen_Norge.jpg
Slide 21
Moment tensor inversion at
Pyhasalmi ore mine:
accuracy test using explosions
Daniela Kühn (NORSAR)
V. Vavrycuk (Academy of Sciences of the CR)
AIM 2nd annual meeting, 29-30 Sept 2011, Prague, Czech Republic
Pyhäsalmi ore mine, Finland
• microseismic monitoring:
since January 2003
safety of the underground
personnel
optimisation of mining
process
Introduction
Waveform
modelling
Moment
tensor
inversion
• network:
12 1-C geophones
+ 6 3-C geophones (ISS)
3-D geometry
sampling rate: < 3000 Hz
Summary
owned by Inmet Mining Co.
• events:
1500 events /months
(including blasting)
-2 < Mw < 1.5
Inconsistent polarities of P-wave first onset
Moment tensor inversion:
• homogeneous velocity model (as for locations)
• amplitude inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling
Complexity of velocity model
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling: 2D section
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
• E3D: viscoelastic 3-D FD code (Larsen and Schultz, 1995)
• strong interaction with mining cavities: reflection,
scattering, conversion
620 m
• healing of wavefronts
Waveform modelling
synthetic seismograms
- complex waveforms
Introduction
- strong coda
- complex secondary arrivals
Waveform
modelling
- scattering effects stronger
on amplitudes than travel
times,
since
size
of
heterogeneities
(cavities,
access tunnels) same order or
smaller than wavelengths
Moment
tensor
inversion
Summary
arrival times computed by
Eikonal solver still fit
(wavefronts heal quickly
after passing a cavitiy)
observed seismograms
Comparison 1-D/3-D
Imaginary
network!
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Comparison 1-D/3-D
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Influence of proximity to cavity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Real mine
network!
Source depth → ray path
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Ray path → onset polarity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Moment tensor inversion
Amplitude picking
direct wave scattered wave
Introduction
Waveform
modelling
Moment
tensor
inversion
first maximum amplitude = amplitude of the direct wave
direct wave
scattered wave
Summary
energy diffracted
around cavity
?
first maximum amplitude is not always the amplitude of the direct wave
Amplitude vs. waveform inversion
Amplitude inversion:
• homogeneous model of the medium
Introduction
• Green’s functions calculated using ray theory
• inversion of P-wave amplitudes (20-30 amplitudes)
Waveform
modelling
• frequencies: 250-500 Hz
cannot take into account distortion of rays on focal sphere
Moment
tensor
inversion
misinterpretation of amplitudes: which one is the direct wave?
Waveform inversion:
Summary
• 3-D heterogeneous model of the medium
• Green’s functions calculated using FD code
• inversion of full waveforms (15-20 waveforms)
• frequencies (at the moment): 25-100 Hz
inversion is performed in frequency domain
in principle same inversion algorithm as for amplitude inversion, but run
repeatedly for every frequency band (0.5 Hz steps)
Selected explosions: relocation
expl 1
expl 1
expl 3
Introduction
expl 5
Waveform
modelling
expl 3
Moment
tensor
inversion
expl 5
expl 1
Summary
expl 3
expl 5
Explosions (coords in m):
1) x=8306E y=2312N z=-1238
→ 26 m shift
3) x=8218E y=2192N z=-1352
→ 68 m shift
5) x=8214E y=2168N z=-1356
→ 59 m shift
Examples: good fit
Examples: amplitude misfit
Examples: phase misfit
Inversion results
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
explosion 1
explosion 3
explosion 5
Length of time windows
GF duration [10 ms]
Data duration [10 ms]
explosion 1
ISO percentage
GF duration [10 ms]
ISO percentage
GF duration [10 ms]
ISO percentage
Data duration [10 ms]
explosion 3
Data duration [10 ms]
explosion 5
nearly all solutions have high isotropic percentage
best solutions near diagonal
(length of GF time window = length of data time window)
best solutions P –wave + S-wave onset
Explosion 1
Amplitude inversion
Waveform inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
DC = 7%
CLVD = -14%
ISO = 79%
Summary: seismicity in mines
structural model in mines is very complex
Introduction
large and abrupt changes in velocity at
cavities
model varies in time
Waveform
modelling
Moment
tensor
inversion
Summary
earthquake source is complex (single
forces, non-DC components)
small changes in source position lead to
large changes in ray propagation, rays can
be strongly curved
radiated wave field is complex
(reflected, converted, scattered waves, head
waves)
Summary: waveform inversion
In general:
• complex Green’s functions can be calculated by 3-D FD codes (accurate model needed!);
Introduction
• sensitive to time shifts due to mislocation or due to inaccurate velocity model
• frequency band of inverted waves can be easily controlled => stability analysis
Waveform
modelling
In particular:
Moment
tensor
inversion
• good network configuration => focal sphere nicely covered
• inversion algorithm:
• optimal with same window length for Green’s functions and data
Summary
• optimal with simultaneous inversion of P- and S-wave, but excluding S-wave coda
• yields high isotropic percentage, higher than amplitude inversion, almost
independently of window length
promising, but computationally demanding (especially the computation of Green’s
functions with sufficiently small grid point distances)
will be performed for selected events, not whole database
Thank you
for your
attention!
http://commons.wikimedia.org/wiki/File:Preikestolen_Norge.jpg
Slide 22
Moment tensor inversion at
Pyhasalmi ore mine:
accuracy test using explosions
Daniela Kühn (NORSAR)
V. Vavrycuk (Academy of Sciences of the CR)
AIM 2nd annual meeting, 29-30 Sept 2011, Prague, Czech Republic
Pyhäsalmi ore mine, Finland
• microseismic monitoring:
since January 2003
safety of the underground
personnel
optimisation of mining
process
Introduction
Waveform
modelling
Moment
tensor
inversion
• network:
12 1-C geophones
+ 6 3-C geophones (ISS)
3-D geometry
sampling rate: < 3000 Hz
Summary
owned by Inmet Mining Co.
• events:
1500 events /months
(including blasting)
-2 < Mw < 1.5
Inconsistent polarities of P-wave first onset
Moment tensor inversion:
• homogeneous velocity model (as for locations)
• amplitude inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling
Complexity of velocity model
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling: 2D section
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
• E3D: viscoelastic 3-D FD code (Larsen and Schultz, 1995)
• strong interaction with mining cavities: reflection,
scattering, conversion
620 m
• healing of wavefronts
Waveform modelling
synthetic seismograms
- complex waveforms
Introduction
- strong coda
- complex secondary arrivals
Waveform
modelling
- scattering effects stronger
on amplitudes than travel
times,
since
size
of
heterogeneities
(cavities,
access tunnels) same order or
smaller than wavelengths
Moment
tensor
inversion
Summary
arrival times computed by
Eikonal solver still fit
(wavefronts heal quickly
after passing a cavitiy)
observed seismograms
Comparison 1-D/3-D
Imaginary
network!
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Comparison 1-D/3-D
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Influence of proximity to cavity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Real mine
network!
Source depth → ray path
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Ray path → onset polarity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Moment tensor inversion
Amplitude picking
direct wave scattered wave
Introduction
Waveform
modelling
Moment
tensor
inversion
first maximum amplitude = amplitude of the direct wave
direct wave
scattered wave
Summary
energy diffracted
around cavity
?
first maximum amplitude is not always the amplitude of the direct wave
Amplitude vs. waveform inversion
Amplitude inversion:
• homogeneous model of the medium
Introduction
• Green’s functions calculated using ray theory
• inversion of P-wave amplitudes (20-30 amplitudes)
Waveform
modelling
• frequencies: 250-500 Hz
cannot take into account distortion of rays on focal sphere
Moment
tensor
inversion
misinterpretation of amplitudes: which one is the direct wave?
Waveform inversion:
Summary
• 3-D heterogeneous model of the medium
• Green’s functions calculated using FD code
• inversion of full waveforms (15-20 waveforms)
• frequencies (at the moment): 25-100 Hz
inversion is performed in frequency domain
in principle same inversion algorithm as for amplitude inversion, but run
repeatedly for every frequency band (0.5 Hz steps)
Selected explosions: relocation
expl 1
expl 1
expl 3
Introduction
expl 5
Waveform
modelling
expl 3
Moment
tensor
inversion
expl 5
expl 1
Summary
expl 3
expl 5
Explosions (coords in m):
1) x=8306E y=2312N z=-1238
→ 26 m shift
3) x=8218E y=2192N z=-1352
→ 68 m shift
5) x=8214E y=2168N z=-1356
→ 59 m shift
Examples: good fit
Examples: amplitude misfit
Examples: phase misfit
Inversion results
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
explosion 1
explosion 3
explosion 5
Length of time windows
GF duration [10 ms]
Data duration [10 ms]
explosion 1
ISO percentage
GF duration [10 ms]
ISO percentage
GF duration [10 ms]
ISO percentage
Data duration [10 ms]
explosion 3
Data duration [10 ms]
explosion 5
nearly all solutions have high isotropic percentage
best solutions near diagonal
(length of GF time window = length of data time window)
best solutions P –wave + S-wave onset
Explosion 1
Amplitude inversion
Waveform inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
DC = 7%
CLVD = -14%
ISO = 79%
Summary: seismicity in mines
structural model in mines is very complex
Introduction
large and abrupt changes in velocity at
cavities
model varies in time
Waveform
modelling
Moment
tensor
inversion
Summary
earthquake source is complex (single
forces, non-DC components)
small changes in source position lead to
large changes in ray propagation, rays can
be strongly curved
radiated wave field is complex
(reflected, converted, scattered waves, head
waves)
Summary: waveform inversion
In general:
• complex Green’s functions can be calculated by 3-D FD codes (accurate model needed!);
Introduction
• sensitive to time shifts due to mislocation or due to inaccurate velocity model
• frequency band of inverted waves can be easily controlled => stability analysis
Waveform
modelling
In particular:
Moment
tensor
inversion
• good network configuration => focal sphere nicely covered
• inversion algorithm:
• optimal with same window length for Green’s functions and data
Summary
• optimal with simultaneous inversion of P- and S-wave, but excluding S-wave coda
• yields high isotropic percentage, higher than amplitude inversion, almost
independently of window length
promising, but computationally demanding (especially the computation of Green’s
functions with sufficiently small grid point distances)
will be performed for selected events, not whole database
Thank you
for your
attention!
http://commons.wikimedia.org/wiki/File:Preikestolen_Norge.jpg
Slide 23
Moment tensor inversion at
Pyhasalmi ore mine:
accuracy test using explosions
Daniela Kühn (NORSAR)
V. Vavrycuk (Academy of Sciences of the CR)
AIM 2nd annual meeting, 29-30 Sept 2011, Prague, Czech Republic
Pyhäsalmi ore mine, Finland
• microseismic monitoring:
since January 2003
safety of the underground
personnel
optimisation of mining
process
Introduction
Waveform
modelling
Moment
tensor
inversion
• network:
12 1-C geophones
+ 6 3-C geophones (ISS)
3-D geometry
sampling rate: < 3000 Hz
Summary
owned by Inmet Mining Co.
• events:
1500 events /months
(including blasting)
-2 < Mw < 1.5
Inconsistent polarities of P-wave first onset
Moment tensor inversion:
• homogeneous velocity model (as for locations)
• amplitude inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling
Complexity of velocity model
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling: 2D section
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
• E3D: viscoelastic 3-D FD code (Larsen and Schultz, 1995)
• strong interaction with mining cavities: reflection,
scattering, conversion
620 m
• healing of wavefronts
Waveform modelling
synthetic seismograms
- complex waveforms
Introduction
- strong coda
- complex secondary arrivals
Waveform
modelling
- scattering effects stronger
on amplitudes than travel
times,
since
size
of
heterogeneities
(cavities,
access tunnels) same order or
smaller than wavelengths
Moment
tensor
inversion
Summary
arrival times computed by
Eikonal solver still fit
(wavefronts heal quickly
after passing a cavitiy)
observed seismograms
Comparison 1-D/3-D
Imaginary
network!
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Comparison 1-D/3-D
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Influence of proximity to cavity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Real mine
network!
Source depth → ray path
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Ray path → onset polarity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Moment tensor inversion
Amplitude picking
direct wave scattered wave
Introduction
Waveform
modelling
Moment
tensor
inversion
first maximum amplitude = amplitude of the direct wave
direct wave
scattered wave
Summary
energy diffracted
around cavity
?
first maximum amplitude is not always the amplitude of the direct wave
Amplitude vs. waveform inversion
Amplitude inversion:
• homogeneous model of the medium
Introduction
• Green’s functions calculated using ray theory
• inversion of P-wave amplitudes (20-30 amplitudes)
Waveform
modelling
• frequencies: 250-500 Hz
cannot take into account distortion of rays on focal sphere
Moment
tensor
inversion
misinterpretation of amplitudes: which one is the direct wave?
Waveform inversion:
Summary
• 3-D heterogeneous model of the medium
• Green’s functions calculated using FD code
• inversion of full waveforms (15-20 waveforms)
• frequencies (at the moment): 25-100 Hz
inversion is performed in frequency domain
in principle same inversion algorithm as for amplitude inversion, but run
repeatedly for every frequency band (0.5 Hz steps)
Selected explosions: relocation
expl 1
expl 1
expl 3
Introduction
expl 5
Waveform
modelling
expl 3
Moment
tensor
inversion
expl 5
expl 1
Summary
expl 3
expl 5
Explosions (coords in m):
1) x=8306E y=2312N z=-1238
→ 26 m shift
3) x=8218E y=2192N z=-1352
→ 68 m shift
5) x=8214E y=2168N z=-1356
→ 59 m shift
Examples: good fit
Examples: amplitude misfit
Examples: phase misfit
Inversion results
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
explosion 1
explosion 3
explosion 5
Length of time windows
GF duration [10 ms]
Data duration [10 ms]
explosion 1
ISO percentage
GF duration [10 ms]
ISO percentage
GF duration [10 ms]
ISO percentage
Data duration [10 ms]
explosion 3
Data duration [10 ms]
explosion 5
nearly all solutions have high isotropic percentage
best solutions near diagonal
(length of GF time window = length of data time window)
best solutions P –wave + S-wave onset
Explosion 1
Amplitude inversion
Waveform inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
DC = 7%
CLVD = -14%
ISO = 79%
Summary: seismicity in mines
structural model in mines is very complex
Introduction
large and abrupt changes in velocity at
cavities
model varies in time
Waveform
modelling
Moment
tensor
inversion
Summary
earthquake source is complex (single
forces, non-DC components)
small changes in source position lead to
large changes in ray propagation, rays can
be strongly curved
radiated wave field is complex
(reflected, converted, scattered waves, head
waves)
Summary: waveform inversion
In general:
• complex Green’s functions can be calculated by 3-D FD codes (accurate model needed!);
Introduction
• sensitive to time shifts due to mislocation or due to inaccurate velocity model
• frequency band of inverted waves can be easily controlled => stability analysis
Waveform
modelling
In particular:
Moment
tensor
inversion
• good network configuration => focal sphere nicely covered
• inversion algorithm:
• optimal with same window length for Green’s functions and data
Summary
• optimal with simultaneous inversion of P- and S-wave, but excluding S-wave coda
• yields high isotropic percentage, higher than amplitude inversion, almost
independently of window length
promising, but computationally demanding (especially the computation of Green’s
functions with sufficiently small grid point distances)
will be performed for selected events, not whole database
Thank you
for your
attention!
http://commons.wikimedia.org/wiki/File:Preikestolen_Norge.jpg
Slide 24
Moment tensor inversion at
Pyhasalmi ore mine:
accuracy test using explosions
Daniela Kühn (NORSAR)
V. Vavrycuk (Academy of Sciences of the CR)
AIM 2nd annual meeting, 29-30 Sept 2011, Prague, Czech Republic
Pyhäsalmi ore mine, Finland
• microseismic monitoring:
since January 2003
safety of the underground
personnel
optimisation of mining
process
Introduction
Waveform
modelling
Moment
tensor
inversion
• network:
12 1-C geophones
+ 6 3-C geophones (ISS)
3-D geometry
sampling rate: < 3000 Hz
Summary
owned by Inmet Mining Co.
• events:
1500 events /months
(including blasting)
-2 < Mw < 1.5
Inconsistent polarities of P-wave first onset
Moment tensor inversion:
• homogeneous velocity model (as for locations)
• amplitude inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling
Complexity of velocity model
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling: 2D section
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
• E3D: viscoelastic 3-D FD code (Larsen and Schultz, 1995)
• strong interaction with mining cavities: reflection,
scattering, conversion
620 m
• healing of wavefronts
Waveform modelling
synthetic seismograms
- complex waveforms
Introduction
- strong coda
- complex secondary arrivals
Waveform
modelling
- scattering effects stronger
on amplitudes than travel
times,
since
size
of
heterogeneities
(cavities,
access tunnels) same order or
smaller than wavelengths
Moment
tensor
inversion
Summary
arrival times computed by
Eikonal solver still fit
(wavefronts heal quickly
after passing a cavitiy)
observed seismograms
Comparison 1-D/3-D
Imaginary
network!
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Comparison 1-D/3-D
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Influence of proximity to cavity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Real mine
network!
Source depth → ray path
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Ray path → onset polarity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Moment tensor inversion
Amplitude picking
direct wave scattered wave
Introduction
Waveform
modelling
Moment
tensor
inversion
first maximum amplitude = amplitude of the direct wave
direct wave
scattered wave
Summary
energy diffracted
around cavity
?
first maximum amplitude is not always the amplitude of the direct wave
Amplitude vs. waveform inversion
Amplitude inversion:
• homogeneous model of the medium
Introduction
• Green’s functions calculated using ray theory
• inversion of P-wave amplitudes (20-30 amplitudes)
Waveform
modelling
• frequencies: 250-500 Hz
cannot take into account distortion of rays on focal sphere
Moment
tensor
inversion
misinterpretation of amplitudes: which one is the direct wave?
Waveform inversion:
Summary
• 3-D heterogeneous model of the medium
• Green’s functions calculated using FD code
• inversion of full waveforms (15-20 waveforms)
• frequencies (at the moment): 25-100 Hz
inversion is performed in frequency domain
in principle same inversion algorithm as for amplitude inversion, but run
repeatedly for every frequency band (0.5 Hz steps)
Selected explosions: relocation
expl 1
expl 1
expl 3
Introduction
expl 5
Waveform
modelling
expl 3
Moment
tensor
inversion
expl 5
expl 1
Summary
expl 3
expl 5
Explosions (coords in m):
1) x=8306E y=2312N z=-1238
→ 26 m shift
3) x=8218E y=2192N z=-1352
→ 68 m shift
5) x=8214E y=2168N z=-1356
→ 59 m shift
Examples: good fit
Examples: amplitude misfit
Examples: phase misfit
Inversion results
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
explosion 1
explosion 3
explosion 5
Length of time windows
GF duration [10 ms]
Data duration [10 ms]
explosion 1
ISO percentage
GF duration [10 ms]
ISO percentage
GF duration [10 ms]
ISO percentage
Data duration [10 ms]
explosion 3
Data duration [10 ms]
explosion 5
nearly all solutions have high isotropic percentage
best solutions near diagonal
(length of GF time window = length of data time window)
best solutions P –wave + S-wave onset
Explosion 1
Amplitude inversion
Waveform inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
DC = 7%
CLVD = -14%
ISO = 79%
Summary: seismicity in mines
structural model in mines is very complex
Introduction
large and abrupt changes in velocity at
cavities
model varies in time
Waveform
modelling
Moment
tensor
inversion
Summary
earthquake source is complex (single
forces, non-DC components)
small changes in source position lead to
large changes in ray propagation, rays can
be strongly curved
radiated wave field is complex
(reflected, converted, scattered waves, head
waves)
Summary: waveform inversion
In general:
• complex Green’s functions can be calculated by 3-D FD codes (accurate model needed!);
Introduction
• sensitive to time shifts due to mislocation or due to inaccurate velocity model
• frequency band of inverted waves can be easily controlled => stability analysis
Waveform
modelling
In particular:
Moment
tensor
inversion
• good network configuration => focal sphere nicely covered
• inversion algorithm:
• optimal with same window length for Green’s functions and data
Summary
• optimal with simultaneous inversion of P- and S-wave, but excluding S-wave coda
• yields high isotropic percentage, higher than amplitude inversion, almost
independently of window length
promising, but computationally demanding (especially the computation of Green’s
functions with sufficiently small grid point distances)
will be performed for selected events, not whole database
Thank you
for your
attention!
http://commons.wikimedia.org/wiki/File:Preikestolen_Norge.jpg
Slide 25
Moment tensor inversion at
Pyhasalmi ore mine:
accuracy test using explosions
Daniela Kühn (NORSAR)
V. Vavrycuk (Academy of Sciences of the CR)
AIM 2nd annual meeting, 29-30 Sept 2011, Prague, Czech Republic
Pyhäsalmi ore mine, Finland
• microseismic monitoring:
since January 2003
safety of the underground
personnel
optimisation of mining
process
Introduction
Waveform
modelling
Moment
tensor
inversion
• network:
12 1-C geophones
+ 6 3-C geophones (ISS)
3-D geometry
sampling rate: < 3000 Hz
Summary
owned by Inmet Mining Co.
• events:
1500 events /months
(including blasting)
-2 < Mw < 1.5
Inconsistent polarities of P-wave first onset
Moment tensor inversion:
• homogeneous velocity model (as for locations)
• amplitude inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling
Complexity of velocity model
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Waveform modelling: 2D section
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
• E3D: viscoelastic 3-D FD code (Larsen and Schultz, 1995)
• strong interaction with mining cavities: reflection,
scattering, conversion
620 m
• healing of wavefronts
Waveform modelling
synthetic seismograms
- complex waveforms
Introduction
- strong coda
- complex secondary arrivals
Waveform
modelling
- scattering effects stronger
on amplitudes than travel
times,
since
size
of
heterogeneities
(cavities,
access tunnels) same order or
smaller than wavelengths
Moment
tensor
inversion
Summary
arrival times computed by
Eikonal solver still fit
(wavefronts heal quickly
after passing a cavitiy)
observed seismograms
Comparison 1-D/3-D
Imaginary
network!
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Comparison 1-D/3-D
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Influence of proximity to cavity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Real mine
network!
Source depth → ray path
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Ray path → onset polarity
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
Moment tensor inversion
Amplitude picking
direct wave scattered wave
Introduction
Waveform
modelling
Moment
tensor
inversion
first maximum amplitude = amplitude of the direct wave
direct wave
scattered wave
Summary
energy diffracted
around cavity
?
first maximum amplitude is not always the amplitude of the direct wave
Amplitude vs. waveform inversion
Amplitude inversion:
• homogeneous model of the medium
Introduction
• Green’s functions calculated using ray theory
• inversion of P-wave amplitudes (20-30 amplitudes)
Waveform
modelling
• frequencies: 250-500 Hz
cannot take into account distortion of rays on focal sphere
Moment
tensor
inversion
misinterpretation of amplitudes: which one is the direct wave?
Waveform inversion:
Summary
• 3-D heterogeneous model of the medium
• Green’s functions calculated using FD code
• inversion of full waveforms (15-20 waveforms)
• frequencies (at the moment): 25-100 Hz
inversion is performed in frequency domain
in principle same inversion algorithm as for amplitude inversion, but run
repeatedly for every frequency band (0.5 Hz steps)
Selected explosions: relocation
expl 1
expl 1
expl 3
Introduction
expl 5
Waveform
modelling
expl 3
Moment
tensor
inversion
expl 5
expl 1
Summary
expl 3
expl 5
Explosions (coords in m):
1) x=8306E y=2312N z=-1238
→ 26 m shift
3) x=8218E y=2192N z=-1352
→ 68 m shift
5) x=8214E y=2168N z=-1356
→ 59 m shift
Examples: good fit
Examples: amplitude misfit
Examples: phase misfit
Inversion results
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
explosion 1
explosion 3
explosion 5
Length of time windows
GF duration [10 ms]
Data duration [10 ms]
explosion 1
ISO percentage
GF duration [10 ms]
ISO percentage
GF duration [10 ms]
ISO percentage
Data duration [10 ms]
explosion 3
Data duration [10 ms]
explosion 5
nearly all solutions have high isotropic percentage
best solutions near diagonal
(length of GF time window = length of data time window)
best solutions P –wave + S-wave onset
Explosion 1
Amplitude inversion
Waveform inversion
Introduction
Waveform
modelling
Moment
tensor
inversion
Summary
DC = 7%
CLVD = -14%
ISO = 79%
Summary: seismicity in mines
structural model in mines is very complex
Introduction
large and abrupt changes in velocity at
cavities
model varies in time
Waveform
modelling
Moment
tensor
inversion
Summary
earthquake source is complex (single
forces, non-DC components)
small changes in source position lead to
large changes in ray propagation, rays can
be strongly curved
radiated wave field is complex
(reflected, converted, scattered waves, head
waves)
Summary: waveform inversion
In general:
• complex Green’s functions can be calculated by 3-D FD codes (accurate model needed!);
Introduction
• sensitive to time shifts due to mislocation or due to inaccurate velocity model
• frequency band of inverted waves can be easily controlled => stability analysis
Waveform
modelling
In particular:
Moment
tensor
inversion
• good network configuration => focal sphere nicely covered
• inversion algorithm:
• optimal with same window length for Green’s functions and data
Summary
• optimal with simultaneous inversion of P- and S-wave, but excluding S-wave coda
• yields high isotropic percentage, higher than amplitude inversion, almost
independently of window length
promising, but computationally demanding (especially the computation of Green’s
functions with sufficiently small grid point distances)
will be performed for selected events, not whole database
Thank you
for your
attention!
http://commons.wikimedia.org/wiki/File:Preikestolen_Norge.jpg