new waveform-based bayesian full moment tensor inversion and … chen gu.pdf · 2016. 5. 20. ·...
TRANSCRIPT
Waveform-based Bayesian Full Moment Tensor Inversion and Uncertainty Quantification for
the Induced Seismicity in Oil/Gas Fields
Chen Gu Graduate Student,
Department of Earth, Atmospheric, and Planetary Sciences
In collabora<on with Prof. Youssef Marzouk and Prof. M. Nafi Toksöz
MIT Earth Resources Laboratory 2016 Annual Founding Members Mee<ng
May 18, 2016
Slide 2 2016 Annual Founding Members Mee<ng
In many oil/gas fields and hydrofracking there are induced earthquakes due to fluid extraction or injection
• Research motivation – Source mechanisms of induced earthquakes and error bars
• Method – Waveform-based Bayesian moment tensor inversion
• Examples – Synthetics and an example from Oman
v1(tb )!
v1(te )!
!
"
#####
$
%
&&&&&
=
G11,1(tb ) G11,2 (tb ) G11,3(tb ) G12,2 (tb ) G12,3(tb ) G13,3(tb )! ! ! ! ! !
G11,1(te ) G11,2 (te ) G11,3(te ) G12,2 (te ) G12,3(te ) G13,3(te )! ! ! ! ! !
$
%
&&&&&
!
"
#####
M11
M12
M13
M22
M23
M33
$
%
&&&&&&&&
!
"
########
P T
Slide 3 2016 Annual Founding Members Mee<ng
Motivation: Understand fracturing mechanisms, map microseismicity, and evaluate hydrofracking
• More high-quality data • In predictable places
Oil/Gas field
Slide 4 2016 Annual Founding Members Mee<ng
Motivation: Understanding earthquake dynamics, hazard assessment, and damage measurement
Note: Earthquakes occur in the same place of Oil/gas fields
Oil fields in Kuwait Induced earthquakes
4.1
4.5
Slide 5 2016 Annual Founding Members Mee<ng
In many oil/gas fields and hydrofracking there are induced earthquakes due to fluid extraction or injection
• Research motivation – Source mechanisms of induced earthquakes and error bars
• Method – Waveform-based Bayesian moment tensor inversion
• Examples – Synthetics and an example from Oman
v1(tb )!
v1(te )!
!
"
#####
$
%
&&&&&
=
G11,1(tb ) G11,2 (tb ) G11,3(tb ) G12,2 (tb ) G12,3(tb ) G13,3(tb )! ! ! ! ! !
G11,1(te ) G11,2 (te ) G11,3(te ) G12,2 (te ) G12,3(te ) G13,3(te )! ! ! ! ! !
$
%
&&&&&
!
"
#####
M11
M12
M13
M22
M23
M33
$
%
&&&&&&&&
!
"
########
P T
Slide 6 2016 Annual Founding Members Mee<ng
Moment Tensors of Earthquakes
• The mechanisms of the greatest majority of tectonic earthquakes can be described by a “Double Couple” – “DC”, corresponding to a shear fracture
• However some events exhibit more complex source mechanism such as volumetric component (ISO) and “Compensated Linear Vector Dipole” – “CLVD”.
• A complete moment tensor including all these components can be written as:
MT =M11 M12 M13
M21 M22 M23
M31 M32 M33
⎡
⎣
⎢⎢⎢⎢
⎤
⎦
⎥⎥⎥⎥
Aki and Richards (1980)
Slide 7 2016 Annual Founding Members Mee<ng
Moment Tensors of Earthquakes
12
0 1 01 0 00 0 0
⎡
⎣
⎢⎢⎢
⎤
⎦
⎥⎥⎥
16
1 0 00 1 00 0 −2
"
#
$$$
%
&
'''
13
1 0 00 1 00 0 1
!
"
###
$
%
&&&
DC CLVD ISO
−1
−0.5
0
VA11
−0.10
0.10.2
VA21
−0.2−0.1
00.1
VA31
−0.2
0
0.2
VA41
0 1 2 3 4 5 6 7 8−0.2
0
0.2
VA51
Time (s)
0
0.5
1
VA11
−0.10
0.10.20.3
VA21
−0.10
0.10.2
VA31
−0.20
0.2
VA41
0 1 2 3 4 5 6 7 8−0.4−0.2
00.20.4
VA51
Time (s)
−0.4−0.2
00.20.40.6
VA11
−0.10
0.10.2
VA21
−0.10
0.10.2
VA31
−0.20
0.20.4
VA41
0 1 2 3 4 5 6 7 8−0.5
00.5
1
VA51
Time (s)
Slide 8 2016 Annual Founding Members Mee<ng
Least Square vs. Bayesian inversion
• Least square method: cannot quantify the uncertainty well • Bayesian inversion method: estimate the probability density
function of moment tensor solutions
Observed Data
Green’s Function: Velocity structure
v1(tb )!
v1(te )!
!
"
#####
$
%
&&&&&
=
G11,1(tb ) G11,2 (tb ) G11,3(tb ) G12,2 (tb ) G12,3(tb ) G13,3(tb )! ! ! ! ! !
G11,1(te ) G11,2 (te ) G11,3(te ) G12,2 (te ) G12,3(te ) G13,3(te )! ! ! ! ! !
$
%
&&&&&
!
"
#####
M11
M12
M13
M22
M23
M33
$
%
&&&&&&&&
!
"
########
Moment Tensor
Slide 9 2016 Annual Founding Members Mee<ng
Waveform-based Bayesian inversion (Adaptive Metropolis MCMC)
Prior
Information we know about the distribution of Mij before we do any inversion Here we use the first P-wave polarity.
Likelihood
How the waveform data are distributed for a given Mij
Posterior
The distribution of Mij solutions given the data
π 0 (Mij )∝1,0,
pol(Mij ) = polobs,
pol(Mij ) ≠ polobs.
⎧⎨⎪
⎩⎪
P(d |Mij )∝Pε (d −G(Mij ),σε2 )
π (Mij | d)∝P(d |Mij )π 0 (Mij )
Slide 10 2016 Annual Founding Members Mee<ng
Waveform-based Bayesian inversion
Polarity
Waveforms
Mij solu<on distribu<on
π 0 (Mij )
P(d |Mij )
π (Mij | d)
Slide 11 2016 Annual Founding Members Mee<ng
Waveform-based Bayesian inversion
Polarity
Waveforms
Mij solu<on distribu<on High
Probability
Low Probability
M1
M2
π 0 (Mij )
P(d |Mij )
π (Mij | d)
Slide 12 2016 Annual Founding Members Mee<ng
Waveform-based Bayesian inversion
Polarity
Waveforms
Mij solu<on distribu<on M1 High
Probability
Low Probability
M2
π 0 (Mij )
P(d |Mij )
π (Mij | d)
Slide 13 2016 Annual Founding Members Mee<ng
Synthetic test: Experiment setup
Step 1: Use a known synthetic source inside a layered structure Step 2: Calculate synthetic seismograms at five stations Step 3: Add 10% Gaussian noise Step 4: Estimate the moment tensor using the “observed” data Step 5: Quantify the uncertainties
05
10
0
5
10
0
1
2
3
4
T1
West to East (km)
T4T5T2
T3
South to North (km)
Depth
(km)
0 1000 2000 3000 4000 5000 6000 7000
0
500
1000
1500
2000
2500
3000
3500
4000Velocity (m/s)
Depth
(m)
VpVs
x = 8.18 km, y = 6.22 km, z = 1.14 km Strike = 305°, Dip = 20°, Rake = 85°DC%: 61, CLVD%: 17, ISO%: 21, α = 10°
NT1
T2
T5T3
T4
0 50 100 150 200 250 300−1
0
1 x 10−5
ST
A1
East Component
0 50 100 150 200 250 300−2
0
2 x 10−5 North Component
0 50 100 150 200 250 300−2
0
2 x 10−5 Vertical Component
0 100 200 300 400−2
0
2 x 10−6
ST
A2
0 100 200 300 400−5
0
5 x 10−6
0 100 200 300 400−5
0
5 x 10−6
0 100 200 300 400−2
0
2 x 10−6
ST
A3
0 100 200 300 400−2
0
2 x 10−6
0 100 200 300 400−2
0
2 x 10−6
0 50 100 150 200 250 300−5
0
5 x 10−6
ST
A4
0 50 100 150 200 250 300−5
0
5 x 10−6
0 50 100 150 200 250 300−1
0
1 x 10−5
0 50 100 150 200 250−5
0
5 x 10−5
ST
A5
Time (Sec)0 50 100 150 200 250
−2
0
2 x 10−5
Time (Sec)0 50 100 150 200 250
−2
0
2 x 10−5
Time (Sec)
Time (s)
Amp. (A
.U.) Blue: Data
Red: Model
Stable calculate uncertainty
P
T
Slide 14 2016 Annual Founding Members Mee<ng
Synthetic test: AM MCMC Bayesian Inversion Parameter M
ij
105 realizations of Mij
M11
M12
M22
M33
M13
M23 Search
Mij
Slide 15 2016 Annual Founding Members Mee<ng
Synthetic test: Uncertainty Quantification
Source Parameters
True Value
Posterior Mean
Standard Deviation
Strike 305° 304° 7°
Dip 20° 19° 3°
Rake 85° 85° 6°
DC% 61.3% 63.4% 7.7%
CLVD% 17.2% 17.7% 5.6%
ISO% 21.5% 18.9% 3.2%
α 10° 10° 4°
P
T
Slide 16 2016 Annual Founding Members Mee<ng
Induced Earthquakes from an Oil/Gas field in Oman
Slide 17 2016 Annual Founding Members Mee<ng
Oman: induced earthquake seismograms
Blue: Data Red: Model
0 1 2 3−0.2
−0.15
−0.1
−0.05
0
0.05
0.1
0.15
VA 1
1
0 1 2 3−0.08
−0.06
−0.04
−0.02
0
0.02
0.04
0.06
0.08
VA 2
1
0 1 2 3−0.2
−0.15
−0.1
−0.05
0
0.05
0.1
0.15
VA 3
10 1 2 3 4−0.08
−0.06
−0.04
−0.02
0
0.02
0.04
0.06
0.08
VA 4
1
0 1 2−2
−1.5
−1
−0.5
0
0.5
1
1.5
2
2.5
VA 5
1Time (s)
Time (s)
Time (s)
Time (s)
Time (s)
Slide 18 2016 Annual Founding Members Mee<ng
Oman: Uncertainty Quantification
Source Parameters
Posterior Mean
Standard Deviation
Strike 53° 5°
Dip 81° 3°
Rake 154° 5°
DC% 70% 6%
CLVD% 7% 6%
ISO% 22% 4%
α (°) 4° 4°
P
T
Slide 19 2016 Annual Founding Members Mee<ng
Summary
Conclusions
• Waveform-based Bayesian inversion can estimate the full moment tensor of seismicity and quantify the uncertainties of source parameters.
Future work • Stress triggering analysis and geodynamic modeling • Hazard assessment and damage measurement
Slide 20 2016 Annual Founding Members Mee<ng
Acknowledgement
• This research was supported by the Kuwait-MIT Center (CNRE).
• We thank Petroleum Development Oman for providing seismic data in Oman.
• We also thank Dr. A. Al-Enezi and Ms. F. Al-Jeri for providing seismic data in Kuwait and assisting in the analysis.
Contact Email: [email protected]
Thank you!
Slide 22 2016 Annual Founding Members Mee<ng
Summary
Conclusions
• Waveform-based Bayesian inversion can estimate the full moment tensor of induced seismicity and quantify the uncertainties of source mechanisms.
Future work • Stress triggering analysis and geodynamic modeling • Hazard assessment and damage measurement