Imaging and Full Waveform Inversion of seismic data from ... Imaging and Full Waveform Inversion of seismic data from the CO 2 gas cloud at Sleipner Espen Birger Raknes, B˝rge Arntsen,

Theory Results Conclusions

Imaging and Full Waveform Inversion ofseismic data from the CO2 gas cloud at

Sleipner

Espen Birger Raknes, Børge Arntsen, and Wiktor Weibull

Norwegian University of Science and Technology (NTNU)Department of Petroleum Engineering & Applied Geophysics

E-mail: [email protected]

ROSE Meeting 2015April 28th 2015


Background

• The applications of full waveform inversion (FWI) onsynthetic and field data the last decade have proved thatFWI is a promising method for parameter model estimation

• The increase in computational power leads to an increasein possible problem sizes and type of wave phenomenaincluded in the modeling and inversion

• Limited number of 3D applications in the literature


Objectives

• Apply three dimensional elastic isotropic FWI on fieldtime-lapse data from the Sleipner area

• Use the inverted elastic models to obtain depth migrationseismic images of the area before and after ten years ofinjection of CO2.

• Investigate the migration path for the injected CO2 gas.


Outline

• Theory• Full waveform inversion• Imaging work flow

• Results• Baseline inversion• Monitor inversion• Seismic images

• Conclusions


A quick overview of full waveform inversion

Overall goal

Find a parameter model from which it is possible to createsynthetic data that is close to some measured data

Define S(m) as the measure between synthetic and measureddata. The FWI is then the problem

arg minm

S(m)

Solved using an iterative method

mk+1 = mk − αkH−1k gk,

mk model at iteration kgk gradient of S(m) at iteration kHk Hessian of S(m) at iteration kαk step length at iteration k

Start point

End point


A quick overview of full waveform inversion

Overall goal

Find a parameter model from which it is possible to createsynthetic data that is close to some measured data

Define S(m) as the measure between synthetic and measureddata. The FWI is then the problem

arg minm

S(m)

Solved using an iterative method

mk+1 = mk − αkH−1k gk,

mk model at iteration kgk gradient of S(m) at iteration kHk Hessian of S(m) at iteration kαk step length at iteration k

Start point

End point


Schematic view of FWI

Initial

modelModeling

Are synthetic

and real data

close enough?

End

model

Gradient

calculation

New model

In parallelSynchronization

yes no


Time-lapse full waveform inversion

Goal

Use full waveform inversion to quantify changes in time for

parameters affecting wave propagation.

May be used

• as monitoring tool during the life-time of a reservoir

• to monitor injection of CO2 in CCS experiments

• quantify amount of injected CO2


Time-lapse full waveform inversion

Goal

Use full waveform inversion to quantify changes in time for

parameters affecting wave propagation.

Challenges

• Need to perform at least two inversions

• The method may introduce artifacts in the time-lapseimages due to for instance

• non-linearity• ill-posedness• data differences• bad repeatability in the time-lapse data


Seismic imaging work flow

Initial modelBaseline data

Baseline FWI

Baseline model Depth migrationMonitor data

Monitor FWI

Monitor model Depth migration




Baseline FWI

Baseline model Depth migration

Monitor data

Monitor FWI





Baseline FWI


Monitor FWI





Baseline FWI


Monitor FWI



Results: Sleipner data details• Baseline dataset: 1994 survey

• 852 shots, 570840 data traces• 1700 m offset

• Monitor dataset: 2006 survey• 1180 shots, 1274400 data traces• 1700 m offset

0 500 1000 1500 2000 2500 3000X (m)

0

1000

2000

3000

4000

5000

6000

7000

8000

Y (m

)

0

20

40

60

80

100

120

140

160

180

# of CMPs

0 500 1000 1500 2000 2500 3000X (m)

0

1000

2000

3000

4000

5000

6000

7000

8000

Y (m

)

0

25

50

75

100

125

150

175

200

225

250

# of CMPs

Fold maps: left: baseline, right: monitor.


Inversion strategy

• Invert for vp, and couple vs and ρ using empirical

relationships

• Invert sequentially using the frequency bands: 6–8Hz,

6–11Hz, 6–15Hz.

• Source signatures are estimated using FWI

• First, invert for baseline data to obtain a baseline elastic

model

• Second, use target-oriented FWI when inverting for the

monitor data


Estimated source signatures

0.0 0.2 0.4 0.6 0.8 1.0Time (s)

0.0 0.2 0.4 0.6 0.8 1.0Time (s)

0.0 0.2 0.4 0.6 0.8 1.0Time (s)

0.0 0.2 0.4 0.6 0.8 1.0Time (s)

0.0 0.2 0.4 0.6 0.8 1.0Time (s)

0.0 0.2 0.4 0.6 0.8 1.0Time (s)

Source signatures: top: baseline dataset, bottom: monitor dataset.

Left: 6–8 Hz, middle 6–11 Hz, right: 6–15 Hz.


Baseline FWI: Vertical slice

2500 3500 4500 5500y (m)

200400600800100012001400

z (m

)

1500 1700 1900 2100 2300vp (m/s)

Initial model


Baseline FWI: Vertical slice

2500 3500 4500 5500y (m)

200400600800100012001400

z (m

)

1500 1700 1900 2100 2300vp (m/s)

Final model


Baseline FWI: Data

−1.0 −0.5 0.0 0.5 1.0Amplitude

0.0

0.5

1.0

1.5

2.0

Tim

e (s

)

InitialFinalField

−1.0 −0.5 0.0 0.5 1.0Amplitude

0.0

0.5

1.0

1.5

2.0

Tim

e (s

)

InitialFinalField

−1.0 −0.5 0.0 0.5 1.0Amplitude

0.0

0.5

1.0

1.5

2.0

Tim

e (s

)

InitialFinalField

Comparison: left: 6–8 Hz, middle: 6–11 Hz, right: 6–15 Hz.


Baseline migration: Vertical slice

2500 3500 4500 5500y (m)

200

400

600

800

1000

1200

1400

z (m

)

Initial model


Baseline migration: Vertical slice

2500 3500 4500 5500y (m)

200

400

600

800

1000

1200

1400

z (m

)

Final inverted model


Baseline FWI: Horizontal slice

400 900 1400 1900x (m)

2200

2900

3600

4300

5000

5700

y (m

)

1700 1800 1900 2000vp (m/s)

400 900 1400 1900x (m)

2200

2900

3600

4300

5000

5700

y (m

)

1700 1800 1900 2000vp (m/s)

vp: left: initial, right: final.



400 900 1400 1900x (m)

2200

2900

3600

4300

5000

5700

y (m

)

400 900 1400 1900x (m)

2200

2900

3600

4300

5000

5700

y (m

)

Initial: left: seismic image, right: overlay.



400 900 1400 1900x (m)

2200

2900

3600

4300

5000

5700

y (m

)

400 900 1400 1900x (m)

2200

2900

3600

4300

5000

5700

y (m

)

Final: left: seismic image, right: overlay.


Monitor FWI: Vertical slice

2500 4000 5500y (m)

200

600

1000

1400

z (m

)

1500 1750 2000 2250 2500vp (m/s)

Baseline



2500 4000 5500y (m)

200

600

1000

1400

z (m

)

1500 1750 2000 2250 2500vp (m/s)

Monitor



2500 4000 5500y (m)

200

600

1000

1400

z (m

)

−400 −200 0 200 400vp (m/s)

Time-lapse


Migration: Vertical slice

3500 4000 4500 5000y (m)

700

800

900

1000

1100

z (m

)

3500 4000 4500 5000y (m)

700

800

900

1000

1100

z (m

)

Close-up of gas cloud: top: baseline, bottom: monitor.


Migration: Horizontal slice

750 1250 1750x (m)

2500

3000

3500

4000

4500

5000

5500

y (m

)

750 1250 1750x (m)

2500

3000

3500

4000

4500

5000

5500

y (m

)

750 1250 1750x (m)

2500

3000

3500

4000

4500

5000

5500

y (m

)Slice at z = 881.25m: left: baseline, middle: monitor, right: overlay



750 1250 1750x (m)

2500

3000

3500

4000

4500

5000

5500

y (m

)

750 1250 1750x (m)

2500

3000

3500

4000

4500

5000

5500

y (m

)

750 1250 1750x (m)

2500

3000

3500

4000

4500

5000

5500

y (m




750 1250 1750x (m)

2500

3000

3500

4000

4500

5000

5500

y (m

)

750 1250 1750x (m)

2500

3000

3500

4000

4500

5000

5500

y (m

)

750 1250 1750x (m)

2500

3000

3500

4000

4500

5000

5500

y (m



Migration: Different migration models

750 1250 1750x (m)

2500

3000

3500

4000

4500

5000

5500y (m

)

750 1250 1750x (m)

2500

3000

3500

4000

4500

5000

5500

y (m

)

Slice at z = 918.75m: left: baseline FWI model, right: monitor FWI model


Conclusions

• FWI improves the elastic models and matches the field data

• FWI produces models that can improve the resolution and

focusing of seismic images

• Injected gas at Sleipner has migrated into structures that

are visible on the baseline images

• The gas has migrated upwards through a fault


Acknowledgements

We thank the BIGCCS centre and the ROSE consortium for

financing this research.

Documents

Imaging and Full Waveform Inversion of seismic data from ... Imaging and Full Waveform Inversion of seismic data from the CO 2 gas cloud at Sleipner Espen Birger Raknes, B˝rge Arntsen,