You are on page 1of 62

Lithology Prediction using

Seismic Inversion Attributes

Dan Hampson
April 22, 2010
Overview

A new product will be released by Hampson-Russell, called


LithoSI.

LithoSI is used to predict lithology or classified well logs from


pre-stack inversion attributes.

This talk presents the theory behind LithoSI, followed by two


examples.

April, 2010 2
The basic problem

It is now routine to
P-Impedance
invert pre-stack data.

S-Impedance

Vp/Vs Ratio

That process transforms


angle gathers into elastic
volumes.
April, 2010 3
The basic problem

The challenge is to P-Impedance


relate the derived
volumes back to the
well logs. S-Impedance

Vp/Vs Ratio

April, 2010 4
The basic problem

One way to do this is to cross plot inversion attributes from the logs:

Vp/Vs Ratio

P-Impedance
April, 2010 5
The basic problem

We can then highlight selected zones on


the cross plot:

Vp/Vs Ratio

Gas sand

P-Impedance
April, 2010 6
The basic problem

The selected zone is plotted on the


inversion volume:

Gas sand

InvertedVp/Vs Ratio Interpreted Gas Sand

April, 2010 7
The basic problem

The problem is that the interpreted


area depends critically on the size of
the zone:

Gas sand

InvertedVp/Vs Ratio Interpreted Gas Sand

April, 2010 8
The basic problem

The problem is that the interpreted


area depends critically on the size of
the zone:

Gas sand

InvertedVp/Vs Ratio Interpreted Gas Sand

April, 2010 9
The basic problem
Instead of defining an arbitrary sharp
Vp/Vs Ratio
cutoff for the zone, we can define a
probability distribution:

P-Impedance

Gas sand

April, 2010 10
The basic problem

This probability distribution is mapped


to the inversion volumes using
Bayesian Classification:

April, 2010 11
The basic problem

How to improve the basic cross plot zone method:

(1) Select multiple zones automatically using lithology logs.

(2) Generate probability distributions for each zone.

(3) Use cross-validation to check zones with seismic.

(4) When applying to the volume, calculate not only the best
zone, but also the probability of occurrence.

April, 2010 12
LithoSI Facies Classification
Lithology Zp Vp/Vs Training set
2000 4000 60001.8 2.2 2.6
Vp/Vs 1600 1700 1800 1900 2000 2100 2200 2300 2400 1600 1700 1800 1900 2000 2100 2200 2300 2400

2.5 2.5
1200 2.5 2.5
2.4 2.4
1250 2.4 2.4
2.3 2.3
1300 2.3 2.3
2.2 2.2
1350 2.2 2.2
2.1 2.1
1400 2.1 2.1
2.0 2.0
1450 2.0 2.0
1.9 1.9
1500 1.9 1.9
1.8 1.8
1550 1.8 1.8
1600 1600 1700 1800 1900 2000 2100 2200
Zp
2300 2400
1600 1700 1800 1900 2000 2100 2200 2300 2400

Extract cross plot Compute probability


data at well locations. distributions.

Oil
Water

gas

p(Class i I p , PR)
Extract probabilities
for each class at
arbitrary locations.
April, 2010 13
Extracting the cross plot data

Each input well requires a


lithology log in addition to
desired rock properties curves:

Lithology Zp Vp/Vs
1600 2000 2400 1.8 2.2 2.6
1600 1700 1800 1900 2000 2100 2200 2300 2400

1200 2.5 2.5

2.4 2.4
1250
2.3 2.3
1300
2.2 2.2
1350 2.1 2.1

1400 2.0 2.0

1.9 1.9
1450
1.8 1.8
1500
1600 1700 1800 1900 2000 2100 2200 2300 2400
1550

1600

The lithology log defines the zones boundaries for


extracting the cross plot points.

April, 2010 14
Bayes Theorem
Bayes Theorem calculates the probability we have a particular
class, given a particular set of seismic attributes:

p (c ) p ( X | c )
p c| X
p( X )
Where:
c is a class (e.g. Sand).
X is a seismic attributes Vector (e.g. X = (Zp, Vp/Vs)).
p(c) is the a-priori probability for class c (e.g. probability of
getting sand in general).
p(X|c) is the probability of attributes X knowing we are in
class c (e.g. distribution of (Zp, Vp/Vs) in sand).
p(X) is the probability of attributes X.

April, 2010 15
Pdf computation
The cross plot is convolved with a smooth kernel function: p (c ) p ( X | c )
p c| X
p( X )

x
xi (e.g., Zp)

The kernel estimate of a 1-D PDF is given by:


Kernel function K
1 n
x xi
f ( x ) K( )
nh i 1 h
2h
This user parameter controls the smoothing of the distribution
April, 2010 16
Effect of the smoothing parameter

Vp/Vs
Small h

Zp
Large h

April, 2010 17
Probability cube computation
p (c ) p ( X | c )
p c| X
P p( X )

Density functions weighted


by a priori class proportions

X
Normalisation

1
Probabilities:
Class A 0.34
0.10
N/A
Class B 0.22
0.46
N/A
Class C 0.44
N/A
0

x3 x1 x2
April, 2010 18
Confusion matrix

Thanks to Brian Russell and Wikipedia:

April, 2010 19
Confusion matrix

Thanks to Brian Russell and Wikipedia:

Diagonals show successful classification.

April, 2010 20
Confusion matrix

Thanks to Brian Russell and Wikipedia:

Off-diagonals show confusion.

April, 2010 21
Validation using the Confidence matrix
Zp
Lithology Log Classified
Inversion trace

SHALES
HC SAND
SAND

April, 2010 22
Validation using the Confidence matrix
Zp
Lithology Log Classified
Inversion trace How often do we get a HC Sand when
there is actually a HC sand present? 83%

SHALES
HC SAND
SAND

April, 2010 23
Validation using the Confidence matrix
Zp
Lithology Log Classified
Inversion trace How often do we get a shale when there
is actually a sand present? 27%

SHALES
HC SAND
SAND

April, 2010 24
Workflow overview
1600 1700 1800 1900 2000 2100 2200 2300 2400
1600 2000 24001.8 2.2 2.6

Zp 1200
Vp/Vs
2.5

2.4
2.5

2.4
1250

1300 2.3 2.3

1350 2.2 2.2

1400 2.1 2.1

Vp/Vs 1450 2.0 2.0

Zp
1500 1.9 1.9

1550 1.8 1.8

1600 1600 1700 1800 1900 2000 2100 2200 2300 2400

Extraction of training set Compute


from wells multivariate
Input data
PDFs
1600 1700 1800 1900 2000 2100 2200 2300 2400

2.5 2.5
1.0 2.4 2.4
Oil
2.3 2.3
Water
2.2 2.2

2.1 2.1
0.0 gas
2.0 2.0

1.9 1.9

1.8 1.8
p(Class i I p , PR) 1600 1700 1800 1900 2000 2100 2200 2300 2400

Compute probability litho-cubes Extract litho-class probabilities

April, 2010 25
Example: West Africa

Crossline Direction Inline Direction

Hz 1

Hz 2
500 ms

Hz 3
Time
1 km

Deep offshore west Africa


Well3
Complex turbiditic deposit Well3

environment
Well1
35 sqkm Well1

12.5x12.5m bin size 4ms Well2


Well2
Several targets in 800ms Hz 1 time map
time window
April, 2010 26
AVA response and lithology / fluid
discrimination
0.5
GR VP VS RHO AVA VSH SW PHIE

6000

Poissons ratio
100 ms

S Impedance
VSH
SAND
0 1
SHALE
0 P Impedance
3000 11000

SW
0 1

1500 P Impedance
3000 11000

April, 2010 27
Seismic data quality and frequency
bandwidth
Near: 12-20 (2-23/28-60Hz)
Near Far Ultra Far Extracted wavelets
500 ms

Far: 20-36(2-17/25-55 Hz) Near Far Ultra Far

Seismic Power spectra


90
NEAR

FAR
ULTRA
FAR
Ultra far: 36-50(2-12/17-37 Hz)

30
0 Hz 100

April, 2010 28
Simultaneous Elastic Inversion in Stratigraphic
Framework
Structural Framework Initial Layered Model Synthetic Angle Stacks Real Angle Stacks

Initial Model Difference

Simulated Annealing
Model Perturbation < ?
Vp, Vs, and t

Final Model

Final model
April, 2010 29
Initial model for simultaneous inversion

1 Km
100 ms

Vp/Vs
1.6 2.6

Property filling of the Vp, Vs and density attributes performed by kriging within
layers of the corresponding 5-10Hz low pass filtered logs
March, 2010 LithoSI Presentation 30
Inverted Vp/Vs attribute

1 Km
100 ms

Vp/Vs
1.6 2.6

Near seismic residuals (Real Seismic Synthetic Seismic) and VpVs attribute
Vp/Vs
April, 2010 31
1.6 2.6
Inversion QC and validation at well 1

Seismic wavelets : 12-20 20-36 36-60


0 VSH 1 VP/VS 5000 IP 10000 IS 0 0 0 0 SW 1
100 ms

1.4 2.8 2000 6000


Well log Real seismic trace
Initial model Synthetic trace from well
Final inverted model Synthetic trace from inversion results
April, 2010 32
Training Set Computation at Different Scales
Inverted attributes
Logs Upscaled Logs extracted near wells

11x11 traces
Litho Zp PR
Poisson Ratio

Poisson Ratio

P-Impedance P-Impedance Poisson Ratio


P-Impedance
Effect of Smoothing Parameter
0.5

h=3 h= 6 h= 10

Poissons Ratio

Poissons Ratio
0
3000 11000
P impedance
Kernel function K

2h
April, 2010 34
Kernel Density Estimate for Lithology
Prediction
VP VS RHO IP IS SW VSH PR VP/VS
0.5
100 ms

Poisson Ratio

SHALE 0
OIL-SAND 3000 P impedance 11000
April, 2010 35
WATER-SAND
Kernel Density Estimate for Lithology
Prediction
VP VS RHO IP IS SW VSH PR VP/VS
0.5
100 ms

Poisson Ratio
p(Z p , PR oil sand )
SHALE 0
OIL-SAND 3000 P impedance 11000
April, 2010 36
WATER-SAND
Validation: Input data
VP VS RHO IP IS SW VSH PR VP/VS
100 ms

SHALES Logs colored by Litho log


HC SAND
SAND
April, 2010 37
Validation: Classification result at
Well Locations
VP VS RHO IP IS SW VSH PR VP/VS
100 ms

SHALES Logs colored by most probable facies derived


HC SAND from seismic attribute at well location
SAND
April, 2010 38
Litho Probability Cubes
Probability of Shales Probability of Water Sand
1 Km
100 ms

0 1 0 1

VSH SW
Probability of HC Sand Most Probable Facies

SHALES
HC SAND
0 1 SAND
LITHO log

April, 2010 39
3-D Visualization of HC Sands

p( HC Sand I p , PR) 80%

Vp/Vs
1.6 2.6 HC sand probability
0 1

April, 2010 40
Example 2: Brazeau area, Alberta

2005 study
2003 study

Brazeau
River

Brazeau
South

2007 study
Primary Targets:
Clastics
Belly River, Cardium,
Viking
Carbonates
Elkton/Shunda, Nisku

April, 2010 41
Example 2 dataset
Area - ~100 km2

1 relocated well (A) with a log set on the full


inversion window, used for the low frequency
initial model.

2 productive gas wells (P,E), used for the inversion


QC and the lithological interpretation.

April, 2010 42
Petrophysical Analysis - Viking

SP DRHO WIRE.PHIE_RHOG_1 PHIE_RHOG_1


-160 MV 40 -400 K/M3 100 0.5 V/V 0 1 V/V 0

0
GR_1
GAPI 200 1950
DEN_1
K/M3 2950 0
WIRE.BADHOLE_1
10 0.2
ILD_1
OHMM 2000 0
COAL_1
V/V 1 Viking Sand Oil (Class I)
150
CALI_1
MM 400
DEPTH
METRES 0.45
NPHI
V/V
TOPS
-0.15 1950
WIRE.DEN_4
K/M3 2950 0.2
ILM
OHMM 2000 0
VOL_QUARTZ_6
V/V QUARTZ 1 Gas (Class IIp)
BS DT_1 WIRE.DEN_1 SFL_1 VSH_2
150 MM 400 500 US/M 100 1950 K/M3 2950 0.2 OHMM 2000 0 V/V 1

2519.8

2525
LCRET

2530

2535

VIK
2540

2545

2550

VIKSND
2555

2560

2565 Angle
VIKSSB
JFOU
2570

2576.0 MANN

April, 2010 43
CDP Stack through well locations
Well P Well E

Viking

April, 2010 44
5 degree Angle Stack

Angle Stack Synthetic Angle Stack

April, 2010 45
12 degree Angle Stack

Angle Stack Synthetic Angle Stack

April, 2010 46
20 degree Angle Stack

Angle Stack Synthetic Angle Stack

April, 2010 47
28 degree Angle Stack

Angle Stack Synthetic Angle Stack

April, 2010 48
36 degree Angle Stack

Angle Stack Synthetic Angle Stack

April, 2010 49
Inversion Results Well-P Location
- GR Well Log - IP Well Log - IS Well Log - Rhob Well Log - Seismic Trace - Seismic Trace - Seismic Trace - Seismic Trace - Seismic Trace
- upscaled Log - upscaled Log - upscaled Log - Inverted - Inverted - Inverted - Inverted - Inverted
Synthetic Synthetic Synthetic Synthetic Synthetic
- Inverted IP - Inverted IS - Inverted Rhob
ANGLE 05 ANGLE 12 ANGLE 20 ANGLE 28 ANGLE 36

Viking

April, 2010 50
Inversion Results Well-E Location
- GR Well Log - IP Well Log - IS Well Log - Rhob Well Log - Seismic Trace - Seismic Trace - Seismic Trace - Seismic Trace - Seismic Trace
- upscaled Log - upscaled Log - upscaled Log - Inverted - Inverted - Inverted - Inverted - Inverted
Synthetic Synthetic Synthetic Synthetic Synthetic
- Inverted IP - Inverted IS - Inverted Rhob
ANGLE 05 ANGLE 12 ANGLE 20 ANGLE 28 ANGLE 36

Viking

April, 2010 51
Section through the wells - Initial Model Zp
High
SE P E NW

Viking

Nordegg

Wabamun
Nisku
Ireton

Low
April, 2010 52
Section through the wells Elastic Inversion Zp
High
SE P E NW

Viking

Nisku

Low
April, 2010 53
Section through the wells Elastic Inversion
Density
High
SE P E NW

Viking

Nisku

Low
April, 2010 54
LithoLogs computation
VSH RHOB GR LithoLogs XPLOT QC
A E P

GR

VSH

April, 2010 55
3D Xplot Wells points Viking Mannville
interval

Rho

Vp

Vs

Rho

Vp Vs Sand:
Shale:
April, 2010 56
Distribution Functions computation

Validation at wells:
E P

Rho

Vp

Vs

Rho

Vp Vs Sand:
Shale:
April, 2010 57
Section through the wells
P E 1 P E

Carb
Viking

Shale
Sand
0

E E
P P
Sand
Probability
0 1

April, 2010 Sand probability Most probable facies 58


3D Visualization - VIKING Horizon

April, 2010 59
3D Visualization of Sand Probability

Sand
Probability
0.2 1

April, 2010 60
CE9 Release of LithoSI

LithoSI is
scheduled to
be released
with the CE9
update in
Q4/2010.

April, 2010 61
Summary

A new product will be released by Hampson-Russell, called


LithoSI.

LithoSI is used to predict lithology or classified well logs from


pre-stack seismic inversion attributes.

LithoSI will be released as a part of the general HRS-9


release in early 2011.

This talk has presented the theory behind LithoSI, followed by


two examples.

April, 2010 62

Related Interests