Upload
others
View
2
Download
0
Embed Size (px)
Citation preview
Pcube+ - high resolution horizon update by prestack inversion Øyvind M. Skjæveland and Richard William Metcalfe, Statoil. Odd Kolbjørnsen, Lundin (formerly NR). Per Røe and Ragnar Hauge, Norwegian Computing Centre (NR).
Outline
• Motivation
• Simplified example – thin gas sand – achieving a detailed interpretation
− Classical detuning
− Rock physics inversion walk-through
• Real case with several layers in tuning (Statfjord East flank)
• Concluding remarks
2
Horizon interpretation - customary practice:
Pick one substack.
Interpret horizons in max peak and max trough.
Thin layers - tuning and AVO issues:
Real layer boundaries not in max peak or max trough
Amplitudes and AVO carry information – not used.
Horizons update by prestack inversion:
Using rock physics knowlegde.
Moving horizons away from peaks and troughs.
Quantifying the uncertainty.
Resulting in:
Better volumetrics.
More accurate well placement.
Better understanding of uncertainty.
Horizon consistency check by prestack data.
Motivation – horizon update by inversion Horizons are widely used
• Volumetrics
• Well prognoses
• Geomodels
One of the most important
deliveries from geophysics.
Vp/V
s
AI
3
Simple case - Thin gas sand in tuning How to achieve a more detailed horizon placement
4
Well
gas
5
Below 0 and λ/2 the amplitude is affected by thickness
Well
position
Input horizons
Output horizons
True thick.
= f(amp, int.thick.)
The quick and easy way – detuning
Int.thick. True thick.
amp
True thickness
Knowledge of properties and assumption of blocky sand
enables detailed horizon prediction below tuning thickness
Simple case - Rock physics inversion AI and near stack only. 3 possible Lithology fluid classes (LFC’s)
6
4ms
LFC
grid
Elastic
properties and
wavelet known
from well.
Young shale
AI = 8000
Old shale
AI = 8000
Gas sand
AI = 5000
wavelet
Model 1 – very poor match
7
AI
Synthetic
Young shale
AI = 8000
Old shale
AI = 8000
Gas sand
AI = 5000
wavelet
LFC
configuration
Model 2 – very poor match
8
AI
Synthetic
Young shale
AI = 8000
Old shale
AI = 8000
Gas sand
AI = 5000
wavelet
LFC
configuration
Model 3 poor match
9
AI
Synthetic
Young shale
AI = 8000
Old shale
AI = 8000
Gas sand
AI = 5000
wavelet
LFC
configuration
Model 4 good match
10
AI
Synthetic
Young shale
AI = 8000
Old shale
AI = 8000
Gas sand
AI = 5000
wavelet
LFC
configuration
11
Some good-match models P(y.s) P(o.s.)
Young shale
AI = 8000
Old shale
AI = 8000
Gas sand
AI = 5000
wavelet
Probability: match-weighted sum of all models
P(gas)
Probability: match-weighted sum of all models
12
Some good-match models P(all)
Young shale
AI = 8000
Old shale
AI = 8000
Gas sand
AI = 5000
wavelet
Buland et al 2008
Pcube
0.3 0.4 0.3
13
Old shale
over
young shale
Old shale
over
gas sand
Gas sand
over
young shale
Extra:
Illegal
combinations
Young shale
AI = 8000
Old shale
AI = 8000
Gas sand
AI = 5000
wavelet
Probability: match-weighted sum of all models
Some good-match models P(all)
Buland et al 2008
Pcube
14
Young shale
AI = 8000
Old shale
AI = 8000
Gas sand
AI = 5000
wavelet
Old shale
over
young shale
Old shale
over
gas sand
Gas sand
over
young shale
Illegal
combinations
Probability: match-weighted sum of all legal models
Pcube+
Illegal combinations
impose blocky sand
Buland et al 2008
Pcube
Gas
sand
Brine
sand
Zone 1
Zone 2
Zone 3
Actual setup – prestack case
15
Smoothed model
=Prior model
Inversion
result
wavelets
AI
Vp/V
s
AI
Angle stacks
Old
shale
Young
shale
Gas
sand
Concept
Input horizons Inp
ut
ho
rizo
n u
nce
rtain
ty
Output horizons
w/uncertainty
Inve
rsio
n
Illegal
combinations
-per zone
Knowledge of properties and assumption of blocky sand
enables detailed horizon prediction below tuning thickness
Statfjord East flank
Brent Gp.
Statfjord Fm.
Dunlin Gp.
East flank
slump block C
SFC
SFB
SFA
Prod. start
1979
300+ wells
Partners:
Statoil
Centrica
ExxonMobil
Draupne
Mime
Cook/Amundsen
shale
Shetland
Sand
Drake/Burton
shale
AI
Vp/v
s
LFC properties
Shetland
Mime
Draupne
Reworked Brent
Intra-reservoir Dunlin
Isolated Brent
Dunlin Group
BCU
TIB
BCU + 35ms
TIB + 30ms
(BIB)
BCU – 5 ms
BCU + 10ms
Statfjord – model setup
BCU
T IB
17
Shetland
Mime
Draupne
Reworked Brent
Intra-reservoir Dunlin
Isolated Brent
Dunlin Group
BCU
TIB
BCU + 35ms
TIB + 30ms
(BIB)
BCU – 5 ms
BCU + 10ms
Statfjord – model setup
Seismic section with interpretation
18
Shetland
Mime
Draupne
Reworked Brent
Intra-reservoir Dunlin
Isolated Brent
Dunlin Group
BCU
TIB
BCU + 35ms
TIB + 30ms
(BIB)
BCU – 5 ms
BCU + 10ms
Seismic section with interpretation
and slave horizons
Statfjord – model setup – slave horizons
19
Shetland
Mime
Draupne
Reworked Brent
Intra-reservoir Dunlin
Isolated Brent
Dunlin Group
BCU
TIB
BCU + 35ms
TIB + 30ms
(BIB)
BCU – 5 ms
BCU + 10ms
Seismic section with interpretation
and sand concept model
Statfjord – model setup – slave horizons
20
Shetland
Isolated Brent
Dunlin Group
Shetland
Mime
Draupne
Reworked Brent
Intra-reservoir Dunlin
Isolated Brent
Dunlin Group
BCU
TIB
BCU + 35ms
TIB + 30ms
(BIB)
BCU – 5 ms
BCU + 10ms
BCU ± 10ms
TIB ± 25ms
BIB ± 25ms
Sand smoothed model section
= prior distribution for sand
Probability 0 1
21
Statfjord – model setup – sand prior section
BCU ± 10ms
TIB ± 25ms
BIB ± 25ms
Sand smoothed model section
= prior distribution for sand
Probability 0 1
22
Statfjord – model setup – sand prior section
BCU ± 4ms
TIB ± 11ms
BIB ± 8ms
Posterieor
uncertainty
(2σ ≈ P(95))
Sand inversion result section
= posterior distribution for sand
Probability 0 1
23
Statfjord – model setup – sand post. section
Inversion process – one trace
24
Isolated Brent
TIB ± 11ms
TIB
Deterministic
model
TIB ± 25ms
Human estimate
of uncertainty
Updated to match
prestack data
inversion
near, input horizons
far, input and slave horizons
sand prior model, input horizons
sand posterior, output horizons
far, output horizons
A B C
A B C
A B C
A B C
A B C
A B C
Input data, sand probability and horizons
25 m
i
Mid stack also used as input but not shown
Probability 0 1
Concept model from horizons
250m Probability 0 1
A
B
C
Input (prior
expected
isochrone)
(constant 30ms) A B C
A B C
Isochrone Isolated Brent sand
A
B
C
Output
(posterior
expected
isochrone)
0 65 Isolated Brent thickness, ms
40 20
sand prior model, input horizons
sand posterior, output horizons
26
500m
0 65 Isolated Brent thickness, ms
40 20
A
B
C
A
B
C
A B C
Isochrone prediction vs well result
0 50 Intra-Reservoir Dunlin thickness, ms
30 20 10 40
Formation TVT est.
pre-inv (m)
TVT est.
post-inv(m)
TVT well
obs. (m)
Mime 3 4 4
Draupne 9 10 12
Reworked
Brent
20 18 8
Intra-res
Dunlin
17 7 7
Isolated
Brent
36 58 55
Intra-reservoir
Dunlin shale
expected
thickness
Isolated Brent
sand
expected
thickness
Well B result (Jan 2016)
27
500m
A
B
C
A
B
C
A
B
C
Mean
(P50)
Maximum
(P10)
Minimum
(P90)
Isochrone Isolated Brent sand - uncertainty
0 65 Isolated Brent thickness, ms
40 20 28
500m
Concluding remarks
• Working with sand probability and uncertain horizons as input and output:
− Facilitates integration of geoscience.
− Easy to relate to for all disciplines compared to AI and vp/vs input/output
• Accurate horizon placement - below tuning, away from peak and trough is valuable
− Horizons with uncertainty are often what we want
− Volumetrics, well planning, geomodels, instant isochrones
• Constraining the inversion is key to achieving sub-tuning detail
− Limited number of lithologies – limited combinations of vp,vs,ρ allowed
− Limiting where the lithologies can be – based on geological input
− Excluding non-geological and non-physical layering (e.g. brine just above gas)
• Making amplitudes move horizons is making amplitudes matter.
29
Thanks
• to the Statfjord partnership for the permission to show the Statfjord results.
• to co-authors for
− doing the Statfjord work (Richard)
− developing the technique and setting up a JIP to take it further (NR)
• to you for listening
30