Вы находитесь на странице: 1из 309

STRATA:

Seismic Inversion Workshop


STRATA Course Outline
Overview of Post-stack Inversion
Exercise 1 - Erskine 3D – Initializing
Model Building 1: Log Correlation
Exercise 2 - Erskine 3D - Log Correlation and inversion
Model Building 2: Log Interpolation
Wavelet Extraction
Exercise 3 - Blackfoot - Starting the Project
Model-based Inversion Parameters
Exercise 4 - Blackfoot – Model-based Inversion
Other Inversion Parameters
Exercise 5 - Blackfoot – Other Inversion Methods
Appendix: Overview of Pre-stack Inversion
Exercise 6 – Simultaneous inversion of pre-stack data

February, 2011 2
General Inversion Comments

Inversion is the process of extracting, from seismic data, the


underlying geology which gave rise to that seismic.

Traditionally, inversion has been applied to post-stack seismic data,


with the aim of extracting acoustic impedance volumes (Strata).

Recently, inversion has been extended to pre-stack seismic data, with


the aim of extracting both acoustic and shear impedance volumes.
This allows the calculation of pore fluids (Strata + AVO).

Another recent development is to use inversion results to directly


predict lithologic parameters such as porosity and water saturation
(Emerge).

February, 2011 3
General Inversion Comments
Input Seismic

Post-stack seismic
inversion
transforms an input
seismic volume into
a volume of
acoustic impedance.

Acoustic Impedance
This output display shows
3 components:
(1) Derived AI (colour)
(2) Derived AI (wiggle)
(3) Real AI logs

February, 2011 4
Types of Inversion
These inversion methods are available in STRATA:

Post-stack:

Recursive: Traditional bandlimited inversion


Model Based: Iteratively updates a layered initial model
Sparse Spike: Constrained to produce few events
Colored: Modern derivative of Recursive Inversion

Pre-stack:

Elastic Impedance: Enhancement for pre-stack data


Independent Inversion: Enhancement for pre-stack data
Lambda-mu-rho (LMR): Enhancement for pre-stack data
Simultaneous Inversion: Enhancement for pre-stack data

February, 2011 5
General Forward Model for Inversion

The common forward model for all inversions:

Wavelet

Impedance Reflectivity Seismic


Acoustic
Shear Elastic

February, 2011 6
Inverse Model

Inversion tries to reverse the forward model:

Inverse
Wavelet

Seismic Reflectivity Impedance


Acoustic
Shear Elastic

February, 2011 7
General Forward Model for Inversion

Impedance Reflectivity

Acoustic Impedance
Z i 1  Z i or
Ri  Z= Shear Impedance
Zi
Z i 1  Z i or Ri
Elastic Impedance Zi+1

Acoustic Impedance = VP


Shear Impedance = VS
Elastic Impedance = Complicated formula (later)

February, 2011 8
General Forward Model for Inversion

Reflectivity Seismic

S  W * R  Noise Seismic = Wavelet convolved


with Reflectivity plus noise.

Notes

(1) There are no multiples modeled.


(2) Transmission loss and geometric spreading are ignored.
(3) Frequency-dependent absorption is ignored.
(4) The wavelet may be time varying.

February, 2011 9
General Forward Model for Inversion

The effect of convolving


the wavelet with the
reflectivity is to remove
much of the high-
frequency detail:

February, 2011 10
General Forward Model for Inversion

Convolution in the time domain is


multiplication in the frequency
domain.

As we can see from these plots, the


effect of the wavelet is to remove
both high and low frequencies from
the trace spectrum.

Theoretically, inversion attempts to


recover these lost regions.

February, 2011 11
Non-Uniqueness in Inversion

All inversion algorithms suffer from


“non-uniqueness”.

There is more than one possible


geological model consistent with the
seismic data. The only way to decide
between the possibilities is to use other
information, not present in the seismic
data.

This other information is usually


provided in two ways:
• the initial guess model
• constraints on how far the final
result may deviate from the initial
guess

The final result always depends on the “other


information” as well as the seismic data.

February, 2011 12
Non-Uniqueness in Inversion

Seismic Initial Model

Inversion

February, 2011 13
Bandlimited (Recursive) Inversion

Recursive Inversion, also called Bandlimited Inversion, is the simplest and


earliest form of inversion.

Z i 1  Z i
Ri 
Starting from the
definition of reflection
coefficient: Z i 1  Z i
Zi
Ri
The impedance of the ith
1  Ri Zi+1
+ 1 layer can be Z i 1  Z i
determined from the ith 1  Ri
layer:

Starting at the first layer,


n 1
1  Ri

the impedance of each
successive layer is Z n  Z1
determined by recursively
i 1
1  Ri
applying this formula:
February, 2011 14
Bandlimited Inversion
Z (m/s*g/cc)
In this simple example:

(a) shows that we can recover Z1 = 1000


the true value of impedance if
we have a single spike, but
Z2 = 1500
(b) shows that if we convolve
the spike with a wavelet we
cannot recover the correct value
of impedance.

Z1 = 1000

Z2 = 818
Z3 = 1227
Z4 = 1004

February, 2011 15
Bandlimited Inversion

Step 1:
The initial background model for Recursive Inversion is formed by
filtering an impedance log from a well:

10-Hz
High Cut

February, 2011 16
Bandlimited Inversion

Step 2:
The recursive equation is applied to the seismic trace. (Note: this is
almost identical to a -90 degree phase rotation):

1  ri
Zi  1  Zi *
1- ri

February, 2011 17
Bandlimited Inversion

Step 3:
Add the scaled inversion trace to the filtered model to get the final result:

+ =

February, 2011 18
Bandlimited Inversion

Input Seismic
Recursive Inversion
produces a result
which is bandlimited
to the same
frequency range as
the input seismic
data.

Note the loss of high


frequency detail, as Recursive Inversion
compared with the
well logs.

February, 2011 19
Bandlimited (Recursive) Inversion

Issues in Recursive Inversion:

(1) The wavelet is ignored. This means that the input seismic data
must be zero phase. STRATA automatically “dephases” the data
if an extracted wavelet is available.

(2) Even if the seismic is zero-phase, side-lobes from the actual


wavelet will be interpreted by the algorithm as lithologic
variations.

(3) The inversion result is bandlimited to the frequency range of the


seismic data.

(4) The scaling of the seismic trace to reflectivity is critical to get the
proper range of impedance changes.

February, 2011 20
Model Based Inversion

Model Based Inversion starts with the equation for the convolutional
model:

S  W * R  Noise
Assume that the seismic trace, S, and the wavelet, W, are known.

Assume that the Noise is random and uncorrelated with the signal.

Solve for the reflectivity, R, which satisfies this equation. This is


actually a non-linear problem, so the solution is done iteratively.

February, 2011 21
Model Based Inversion

Step 1:
The initial background model for Model Based Inversion is formed by
blocking an impedance log from a well:

The user specifies the layer size


in milliseconds.

All the layers are originally set


to the same size (in time).

February, 2011 22
Model Based Inversion

Step 2:
Using the blocked model, and the known wavelet, a synthetic trace is
calculated.
Synthetic Seismic

This is compared with the


actual seismic trace.

By analyzing the errors or


“misfit” between synthetic
and real trace, each of the
layers is modified in
thickness and amplitude to
reduce the error.

This is repeated through a


series of iterations.

February, 2011 23
Model Based Inversion

Input Seismic
Model Based
Inversion produces
a broad-band, high
frequency result.

A potential problem
is that the high
frequency detail
may be coming from
the initial guess Model Based Inversion
model, and not from
the seismic data.

This problem is
minimized by using
a smooth initial
model.
February, 2011 24
Model Based Inversion

Recursive Inversion
This is a
comparison
between Recursive
and Model Based
Inversion.

Generally, the
Model Based gives
more detail, but
the results are Model Based Inversion
actually quite
similar.

February, 2011 25
Model Based Inversion
Issues in Model Based Inversion:

(1) Because the wavelet is known, its effects are removed from the
seismic during the calculation. For example, the seismic does not
have to be zero-phase, as long as the wavelet has the same phase
as the seismic.

(2) Errors in the estimated wavelet will cause errors in the inversion
result.

(3) The effective resolution of the seismic is enhanced.

(4) The result can be dependent on the initial guess model. This can be
alleviated by filtering the model.

(5) There is a non-uniqueness problem, as with all inversion.

February, 2011 26
Sparse Spike Inversion

Sparse Spike Inversion


assumes that the actual
reflectivity can be
thought of as a series of
large spikes embedded
in a background of
small spikes.

Sparse Spike Inversion


assumes that only the
large spikes are
meaningful. It finds the
location of these large
spikes by examining the
seismic trace.

February, 2011 27
Sparse Spike Inversion

Sparse Spike Inversion


builds up the
reflectivity sequence
one spike at a time.
Spikes are added until
the trace is modeled
accurately enough.

The amplitudes of the


impedance blocks are
determined using the
Model Based Inversion
algorithm.

February, 2011 28
Sparse Spike Inversion

Input Seismic
Sparse Spike
Inversion
produces a
broad-band, high
frequency result.

Sparse Spike Inversion

February, 2011 29
Sparse Spike Inversion

Model Based Inversion


Sparse Spike
Inversion
produces a result
which is similar to
Model Based
Inversion.

The main
difference is that
the very thin layers Sparse Spike Inversion
are missing.

February, 2011 30
Sparse Spike Inversion (LPSS)

Linear Programming Sparse Spike Inversion seeks the simplest possible


reflectivity model that, when convolved with the wavelet, produces a
synthetic that matches the input seismic.

The simplest model is defined as a model with minimum L1 norm subject


to the constraint that its synthetic matches with the input seismic.

February, 2011 31
Sparse Spike Inversion

Issues in Sparse Spike Inversion:

(1) Sparse Spike Inversion puts events only where the seismic
demands.

(2) It attempts to produce the simplest possible model consistent


with the seismic data.

(3) It often produces fewer events than are known to be geologically


true.

(4) It may be less dependent on the initial guess model than Model
Based Inversion.

February, 2011 32
Colored Inversion

Colored Inversion is a modification of Recursive Inversion, which


was originally described by Lancaster and Whitcombe of BP at the
2000 SEG Convention.

In this process, there is a single operator, O, which is applied to the


seismic trace S to transform it directly into the inversion result Z:

Z  O* S
The authors defined the operator, O, in the frequency domain.

By examining transforms between seismic data and actual inversion


results, they concluded that the operator phase is -90 degrees.

February, 2011 33
Colored Inversion

The amplitude spectrum of the operator is derived this way:

Using a set of wells from Amplitude Spectrum of Acoustic Impedance


the area, the amplitude
spectra of the acoustic
impedance for all the wells

Log(Impedance)
are plotted on a log-log
scale.

As predicted by theory, we
can fit a straight line which
represents the “desired”
output impedance
spectrum.

Log(Frequency)

February, 2011 34
Colored Inversion

Spectrum of Seismic Data


Then, using a set of seismic traces
from around the wells, the average
seismic spectrum is calculated.

From the two preceding spectra, the Operator Spectrum


operator spectrum is calculated. This
has the effect of shaping the seismic
spectrum to the impedance spectrum
within the seismic band.

Frequency (Hz)
February, 2011 35
Colored Inversion

Colored Inversion Operator

Putting together the derived


amplitude spectrum with the -90
degree phase shift produces
the Colored Inversion Operator.

This is applied to all the seismic


traces by convolution.

Time (ms)

February, 2011 36
Colored Inversion

Colored Inversion
Input seismic
produces a result very
similar to Recursive
Inversion.

One difference is that, in


the original
implementation, the
scale is relative Acoustic
Impedance, with positive
and negative values.
Colored Inversion
+3000

February, 2011 -3000 37


The comparison below is the proverbial “apples and
oranges”, since we are comparing absolute to relative AI.

12000
Recursive Inversion

Absolute AI
8300

4600
+3000
Relative Colored Inversion

Relative AI
0

-3000
February, 2011 38
However, we have an option in STRATA to add back
the low frequencies to produce absolute AI, as shown below.

Recursive Inversion 12000

Absolute AI
8300

4600
12000
Absolute Colored Inversion

Absolute AI
8300

4600
February, 2011 39
Colored Inversion
Issues in Colored Inversion:

(1) Very little dependence on the initial model, except to determine the
general impedance trend.

(2) Very fast to apply.

(3) Very simple with few user parameters.

(4) Assumes the data is zero-phase.

(5) Produces a result similar to Recursive Inversion, but with higher


frequency content and better scaling.

(6) In the initial implementation, the method produced a relative


impedance result, although we now have an option to add back the
low frequency trend.

February, 2011 40
AVO Inversion

The basic convolutional model assumes zero-offset data.

Conventional inversion should not be applied to data with AVO


effects, since changes in VP/VS are not explicitly accounted for.

To extend inversion to handle AVO data, these algorithms are


currently used:

(1) Elastic Impedance


(2) Independent Zp and Zs inversion
(3) Simultaneous Inversion for Zp, Zs, and density
(4) Lambda-Mu-Rho (LMR)

These techniques will be discussed later in the course.

February, 2011 41
Stochastic Inversion

Stochastic Inversion

This is a form of geostatistical inversion which explicitly addresses the


non-uniqueness problem by producing a large range of inversion
results for a given input seismic volume.

Each of the results is consistent with the seismic data, and honors the
expected continuity conditions, as contained in the variograms.

These results are analyzed to give an estimate of the uncertainty in the


result, along with the most probable result.

STRATA does not contain a stochastic inversion option.

February, 2011 42
General Inversion Flow

(1) Create Model: Select wells


Correlate each well
Extract wavelet
Import / Pick seismic horizons

(2) Perform Inversion: Select Inversion Type and Parameters


QC Inversion Result

(3) Interpret Result: Create data slices


Create cross plots
Input to EMERGE project

February, 2011 43
QC’ing the Inversion

How do we know the Input seismic


inversion worked?

Two ways:

(1) Error plot


(2) Cross validation

Inversion result

February, 2011 44
QC’ing the Inversion

Input seismic
From the derived
impedance traces, we
can calculate a
synthetic using the
known wavelet.

Ideally, this should


look very much like
the input seismic.

Inversion synthetic

February, 2011 45
QC’ing the Inversion

Input seismic
By subtracting the
Inversion Synthetic from
the Input seismic, we get
the Inversion Error.

If the inversion has


worked well, this should
show very little amplitude
with no localized events.

Because of non- Inversion Error


uniqueness, a small error
does not guarantee the
right answer.

February, 2011 46
QC’ing the Inversion

The second type of inversion QC is cross-validation. In this process, we


drop a well completely from the initial model, perform the inversion at
that location, and compare the result with the hidden well.

Hidden Well

Inversion Error

Inversion Result
February, 2011 47
QC’ing the Inversion

By analyzing the errors at each well location, we can identify problem


wells.

February, 2011 48
Inversion as an Attribute for EMERGE

Inversion A recent use of Inversion is input to


EMERGE, which directly predicts
porosity and other lithologic volumes.

EMERGE

Porosity volume

February, 2011 49
Exercise 1: Erskine 3D, Initializing
The first exercise will apply inversion to a carbonate reef dataset from
central Alberta.

Start the GEOVIEW program by selecting Geoview from Start / Programs /


HRS applications (Windows).

GEOVIEW consists of 2
windows. The first is the
program manager.
The second is the Well
Explorer.

February, 2011 50
We will create a new GEOVIEW database for this project.

When you start the


GEOVIEW program, this
menu appears, allowing
you to open a previously
created database. In this
case, select New and click
on Ok.

February, 2011 51
Name the new database
“erskine_database” as shown
here, and click on Ok:

Now, the Well


Explorer appears
with no wells
entered yet:

February, 2011 52
On the GEOVIEW Well
Explorer window, click on
Import Data / Logs, Check
Shots, Tops, Deviated
Geometry from Files:

On the File Import


page, select the file
“erskine_log.las” and
click on Next >>.

Change the Destination Well


Name field to “erskine_well” and
click on Next >> on this page:
February, 2011 53
Click on Next >> to use the default location information:

This page now appears, showing that there is a single sonic log contained in
the LAS file. Click on Ok to read in this log.

Accept the default display units.

February, 2011 54
After the log is read into GEOVIEW, click on the name of the erskine_well and
then click on Display Well:

The erskine sonic log is displayed.

February, 2011 55
Now we will start a new project in STRATA to perform inversion on the
erskine data set.
Start the STRATA program by clicking on the STRATA
button on the GEOVIEW main window.

Select the option to Start New Project:

Name the new project


“erskine_project”

February, 2011 56
The first step is to read the 3D seismic volume into STRATA. Click on Data
Manager / Import Data / Open Seismic / From SEG-Y File:

Select the file “ersk3d.sgy” and click on Next >>:

February, 2011 57
Once again, we will load the seismic data as a 3D volume.

This file also does not have Inline & Xline numbers or X & Y coordinates in
the trace headers. Change the menu as shown below:

February, 2011 58
Click Next >> twice to accept the defaults until you reach the final page.
The program initially assumes that there is just a single inline.

We will correct the geometry by typing


in “155” as shown. Note that the
number of Inlines will be calculated.

Click on Ok to load the seismic


volume.

February, 2011 59
On the Well To Seismic Map menu, insert the location of the erskine_well
as shown below. Click Ok on this menu.

On the seismic display,


enter “24” (Enter) as the
desired Inline to plot
and the resulting
display will look like
this:

February, 2011 60
To build the initial guess model for this data set, we
need a set of horizons. First, we will pick a single
horizon, and then we will import a set of previously
picked horizons.

Click on Horizon / Pick Horizons:

Accept the default name


“Horizon 1”. Click on Ok. Click
Yes on the dialog that asks if
you would like to display a Map
View.

February, 2011 61
As shown below, change the Mode to Left & Right Repeat. Then, pick the
single horizon shown below by clicking the mouse near it:

The map window shows the pick times


for this single inline.

February, 2011 62
Now we will ask the program to pick the entire 3D volume automatically,
using the single picked inline as a guide. Click on Options / Automatic
Picking:

Click on Ok on the Automatic Picking


menu, and the volume will be picked.

We can see from the pick map that there


is a potential problem on the first couple
of inlines.
February, 2011 63
Display inline 2 by typing that number at the top of the STRATA window
and clicking Enter.

We can see a zone where


the automatic picking has
jumped a leg.

Fix this error manually by clicking near the event:

February, 2011 64
Now do the automatic picking again.

When the automatic picking menu appears, you


can see that the default option is to clear all the
previous automatic picks, and only keep the
manual picks as the new guide:

Click on Ok to get the new result. Note


that picking the second inline manually
improved the model to guide the
automatic picking:

February, 2011 65
Next, we will import the picks. First, delete
the horizon we have just picked:

Then click on Ok on the bottom of the


STRATA window to remove the picking
options.

Click on Horizon /
Import Horizons /
From File:

On the file selection menu,


select the 5 files called
erskine1.pik to erskine5.pik.

February, 2011 66
In this case, the File Type is
Default Geoquest:

Click on Next >> to get the


next page.

We will accept the defaults


on the next page, including
the suggested names and
colors for the horizons.

Click on OK on this menu


to read in the picks.

February, 2011 67
When the picks have been loaded, display inline 24 again, and STRATA
should look like this:

February, 2011 68
Now we will build the initial model for inversion.

Click on Model / Build/Rebuild a Model:

On the model building menus, we will accept all


the defaults. Click Next>> and Ok to create the
model.

(End of Exercise 1)

February, 2011 69
The Initial Guess Model

The initial guess model


for each trace consists
of an impedance log,
usually derived by
multiplying a real sonic
log by a real density log.
The impedance log
model must be
measured in 2-way travel
time. The original logs
are measured in depth.
A critical step is depth-
to-time conversion:

February, 2011 70
The depth-to-time
conversion is made
using a depth-time table
which maps each depth
to the two-way travel
time from the datum
(surface) to that depth
and back:

February, 2011 71
The depth-time table is usually calculated from the sonic log velocities
using this equation:
i
dj
ti  2* 
where: ti = time down to layer i
dj = thickness of layer j
j 1 Vj Vj = velocity of layer j

The time to an event depends


on all the velocities above that
layer, including the first
velocity to the surface, V1.
That velocity is unknown and
is usually approximated by
extrapolating the first
measured velocity back to the
surface:

February, 2011 72
If the well is deviated, it must be corrected to vertical and the correction
made from KB to datum:

DM = Measured depth from KB


DV = Vertical depth from KB
DS = Vertical depth from datum
T = Two-way time from datum

February, 2011 73
The depth-time table calculated from the sonic log is rarely sufficient to
produce a model impedance which ties the seismic data properly because:

 The seismic datum and log datum may be different.

 The average first layer velocity is not known.

 Errors in the sonic log velocities produce cumulative errors in the


calculated travel-times.

 The events on the seismic data may be mispositioned due to


migration errors.

 The seismic data may be subject to time stretch caused by


frequency-dependent absorption and short-period multiples.

February, 2011 74
To improve the depth-time table two procedures are used:

 Apply check shot corrections.


 Apply manual log correlation to the seismic data.

Check Shot Corrections


A check shot table is a series of
measurements of actual 2-way
time for a set of depths:

February, 2011 75
The depth-time
table calculated
from the sonic log
must be modified
to reflect the
desired check shot
times:

Original
Depth/Time
Curve

Desired
Depth/Time
Curve
February, 2011 76
The
interpolation of
points on the
drift curve uses
one of three
options:

Linear: Honors the points exactly with straight line segments between

Spline: Honors the points exactly with smooth curves between

Polynomial: Fits a smooth curve using least-squares optimization


February, 2011 77
Depending on the
interpolation option
used, the sonic log
changes may be
drastic:

Note: The time


stretches in this
example are
unrealistically
large.

February, 2011 78
Changing the depth-time table

Changing the depth-time table implies a


possible change in the original sonic log
velocities. There are three options in STRATA:

(1) Change all the velocities in the log in such


a way that the new log will integrate to exactly
the desired times.

Note: This involves a ramped velocity above


the first measured depth to handle the bulk
time shift and to minimize the effect of
spurious reflections on the synthetic.

This is called “Apply All Changes” in STRATA.

February, 2011 79
(2) Change the velocities for layers between
the first and last check shot depth only.

This means that no ramp is added above the


first measured depth.

The resulting log will integrate to the desired


times except for a bulk time shift.

This is called “Apply Relative Changes” in


STRATA.

February, 2011 80
(3) Do not change the velocities in the sonic
log.

The resulting log will not integrate to the


desired times, but GEOVIEW and STRATA
will use the new depth-time table.

This option has the effect of maintaining the


original reflection coefficients for synthetic
calculations.

This is called “Change Depth-Time Table


Only” in STRATA.

February, 2011 81
Log Correlation

Log correlation is the process of applying a manual correction to the depth-


time curve to optimize the correlation between initial model and seismic
data.

Log correlation should be applied after the check shot correction, and is
ideally a small change.

Log correlation changes the depth-time curve in exactly the same way as a
check shot correction.

Log correlation consists of selecting events on the synthetic trace and the
corresponding events on the real trace.

Since the synthetic is used, the choice of wavelet may be crucial.

February, 2011 82
Exercise 2: Erskine 3D – Log
Correlation and Inversion

Now we are
ready to do log
correlation on
the erskine well.

Click on Well /
Edit/Correlate
Well:

On the selection menu, select


“erskine_well” to correlate:
February, 2011 83
On the eLog window, click on Correlate:

On the Extract Composite Trace menu, accept the default, which is to


extract the composite trace from the ersk3d volume using +/- 1 inline and
cross line:

February, 2011 84
The log correlation window
looks like this:

First, extract a new wavelet. Since


the log has not yet been correlated,
use the Statistical wavelet extraction
to extract a zero-phase wavelet with
the same amplitude spectrum as the
seismic.

February, 2011 85
We will set the analysis
window to use a smaller
Time window and select
traces from a small range of
Inlines and Xlines around
the well:

Use the default values on


the third page of the
Statistical Wavelet
Extraction menu:

February, 2011 86
The extracted wavelet will look like this:

February, 2011 87
Now the Log Correlation window looks like this:
We can see that
there is a mis-tie
between the events
on the synthetic
traces and the
corresponding
events on the real
traces.

We can also see


that the program is
suggesting we
apply a time shift of
14 ms.

To see that better


click on the
Parameters button.
February, 2011 88
The Cross Correlation window
shows the correlation between
the synthetic traces and the
real composite trace.

Note that the maximum


correlation occurs if the
synthetic traces are shifted by
14 ms.

Note, also, that this calculation


can sometimes be improved by
optimizing the Traces
Calculation Window.

For this case, we will leave that


alone.

February, 2011 89
Now go to the Log Correlation window and apply the suggested shift by
clicking on Apply Shift:

Two things happen – first, the logs are shifted; second, the correlation plot is
updated:

February, 2011 90
The shift we have just done is
the best we could do with a
single bulk shift.

To further improve the


correlation, we need to manually
apply time-variant shifts.

To do that, select the series of


points shown on the right by
alternately clicking on the event
on the synthetic (blue) trace and
the corresponding event on the
real (red) trace.

When you have selected the


events as shown, click on the
Stretch button.

February, 2011 91
The default parameters use
Spline interpolation between
points on the drift curve.

February, 2011 92
Change the Type of Interpolation to Linear and click on Apply. Note the
change in the shape of the drift curve.

February, 2011 93
Change the menu as shown below and click on Apply. Note that the option to
Apply all changes adds a ramp to the top of the sonic log, and changes the
sonic log values.

February, 2011 94
Finally, change the menu as
shown below and click on
Apply. Then click on Ok on
the Check Shot window to
accept these parameters.

February, 2011 95
The log correlation window
now looks like this. Note
that we have achieved an
86% correlation..

February, 2011 96
The Cross Correlation window now
shows a strong peak close to time
zero.

Actually, it suggests a further -1ms


time shift. To apply that shift, click
the Apply Shift button once more.

We can also conclude from the


very symmetric correlation shape
that no further phase adjustment is
required.

February, 2011 97
Now that the correlation is done, click on Ok
at the bottom of the eLog window.

The next menu allows you to name the


sonic log that will be created. Click Ok
on this menu to accept the default
name (P-wave_corr).

Finally, click on File / Exit Window on the eLog


window.

February, 2011 98
Running the inversion
Now run the model based inversion using this initial model. We will use the
default parameters and discuss these parameters later.

We will do this in two stages. First


we apply inversion at the well
location to confirm the inversion
parameters and allow the program
to determine the optimum scaling.
Click on Analysis / Post-stack
Analysis / Model Based:

On the first menu page, select ersk3d as the inversion input. Then click on
Next>>

February, 2011 99
On the second page, we confirm that the right wavelet is being used.
Click on Set Current Wavelet to see it.

The display shows our


previously extracted wavelet,
which is right. Click on Cancel
to remove this window.

Then click on Next>> and Ok to


accept all the defaults and
produce the Inversion Analysis
window.

February, 2011 100


The Inversion Analysis
window shows a
number of useful curves
which help confirm that
the inversion has
worked properly. Real Log

Initial
Model

Inversion
Trace

Synthetic Error

Real
Data
February, 2011 101
Many of the display parameters can be
customized by clicking on the “eyeball”
icon.

Select the Curves tab.

And choose the


option to apply a filter
to the real logs.
Then, click on Ok:

February, 2011 102


From this display, we can apply the 2
QC criteria for a good inversion. The
inverted trace (red) corresponds very
closely to the real (filtered) log (blue).
Also, the error or difference between
synthetic (red) traces and real (black)
traces is practically zero.

If we liked, we could
modify any inversion
parameters on the
other menu, and click
Apply to see the new
result. However, this
inversion is definitely
good enough to
proceed.

Click on File / Exit on


the analysis window.
February, 2011 103
Now, we apply inversion to the entire
volume. To do that, click on Inversion
/ Post-stack Inversion / Model Based
Inversion:

On the resulting menu, all the


default parameters are correct,
since we have confirmed them
during the analysis. Also,
clicking Next>> until reaching
the Scaler Option page, we see
that the scalers calculated at
the well location will be used
for the entire volume. Click on
Ok to invert the entire volume.

February, 2011 104


When the inversion of the entire volume has completed, the result will look
like this:

Note that you can move through the volume by clicking the arrow keys as
shown above.
February, 2011 105
One way to evaluate the quality of the inversion result is to create the Error
Plot. This is the difference between the synthetic calculated using the
inversion result and the original data. To see this plot, click on the
“eyeball” icon on the inversion result window.

When the menu appears, set the Trace Data Volume to be the “inverted
derived Synthetic Error”.

Then click on Apply at the bottom of the menu to see the resulting error plot.

February, 2011 106


The Error Plot is scaled at exactly the same scale as the input data. The
fact that there is very little coherent error indicates that the derived model is
a very faithful representation of the seismic data.

Click on File / Exit


Project on any
window to close the
Strata program.

February, 2011 107


A warning message will appear,
asking if you want to save the
project. Click on Yes.

Now a new message will appear,


asking if you want to see a list of
logs which have been modified.
Click on Yes to see that list.

Finally, click on Ok to save the


modified P-wave log back to the
Geoview database, and save the
Strata project as well.

(End of Exercise 2)

February, 2011 108


The Initial Guess Model

Interpolating the Log:

Adding a single log to the model creates a uniform horizontal model:

February, 2011 109


Picking a single event guides the interpolation of the log:

Note: A single picked event simply produces a bulk time shift on the log for
each trace. This is equivalent to applying a check shot correction with a
single point.

February, 2011 110


Picking two or more events is equivalent to applying a variable check-shot
at each trace. The impedances between the two picked events are
stretched / squeezed.

February, 2011 111


The type of interpolation between
horizons is controlled in STRATA by
the Model Option menu:

February, 2011 112


There are 3 options for interpolation as
shown on the right.

By default, all horizons are treated as


Conformable, except the first and the last.

February, 2011 113


A pinch-out is handled by forcing the two picked events to converge:

February, 2011 114


If horizons cross, the handling depends on the
Priority Value assigned to each horizon.

In this case, H1 has a higher priority, so H2 is


truncated.

This is the opposite case.

February, 2011 115


STRATA currently does not handle faults in model building. However, the
effect may be simulated by picking the same event on both sides of the
fault, and picking the fault plane as well:

February, 2011 116


When more than one well is entered into the model, the results are
interpolated using inverse-distance weighting:

February, 2011 117


Assume that there are two input logs, L1 and L2. We wish to calculate the
output log, Lout.

This will be a linear combination of the two input logs:

Lout = w1*L1 + w2*L2

1 d 12
w1 
The weights vary inversely as the
distance from the target point to each of
the input logs: 1 d 12  1 d 2 2
In general:
Lout   wi * Li
-2
i

where: wi  d i

d
-2
j
j
February, 2011 118
The options for inter-well interpolation are shown here:

February, 2011 119


Less frequently used options are
Triangulation, which fits a series of
plane segments between adjacent
wells …

… and Kriging, which


requires a variogram to be
input:

February, 2011 120


Using picked events with multiple logs forces the inverse distance
interpolation to be guided by the picked events:

February, 2011 121


Note the difference between interpolation with and without picked events:

General rules for adding picked events:

(1) Picked events must be present across the entire survey. Missing
picks will be interpolated by the program.
(2) Only pick events which you are sure of.
(3) Pick the large scale structure, not the fine details.
(4) Usually, the events picked during conventional interpretation are
precisely what STRATA needs.
February, 2011 122
By default, STRATA produces a smoothed model by applying a high-cut
frequency filter to the traces after interpolation, maintaining only the low-
frequency trend. This prevents high-frequency details in the model from
influencing the final inversion result.

February, 2011 123


It is also possible
to use the high-
frequency model
that results from
simply interpolating
the model traces,
without any
smoothing.

February, 2011 124


This is a comparison of inversion results from the High Frequency and Smooth
initial models.
High Frequency Model Smooth Model

Inversion Result Inversion Result

February, 2011 125


The results are
surprisingly similar, but
the second is probably
more reliable.

Result from High


Frequency Model

Result from
Smooth Model

February, 2011 126


Wavelet Extraction

The Convolutional Model is used as


the basis for all inversion:

trace = wavelet * reflectivity + noise

In the frequency domain, convolution


becomes multiplication:

Inversion can be thought of as division by the wavelet:

Reflectivity = Trace / Wavelet

The narrow band wavelet restricts the available range of information


in the frequency domain.
February, 2011 127
The wavelet is defined completely by its amplitude spectrum and its phase
spectrum:

Over a limited
frequency range, the
phase spectrum may
often be approximated
by a straight line.

The intercept of the


line is the constant
phase rotation which
best characterizes this
wavelet.

The slope of the line


measures the time-
shift of the wavelet.

February, 2011 128


These wavelets all have the same amplitude 0o
spectrum, but different (constant) phase
spectra:

45o

90o

180o

-90o

February, 2011 129


A special wavelet phase issue is the Polarity Convention.
The default convention is that an increase in acoustic impedance is
represented as a peak on zero-phase seismic data:

The alternate convention is that an increase in acoustic impedance is


represented as a trough on zero-phase seismic data:

The polarity convention is set using the


Synthetic Polarity Convention menu:
February, 2011 130
Wavelets in the earth vary both laterally (spatially) and temporally for a
variety of reasons:

 Near surface effects (space variant)

 Frequency-dependent absorption (space and time variant)

 Inter-bed multiples (space and time variant)

 NMO stretch

 Processing artifacts

STRATA assumes that the wavelet is constant with time and space:

 Time invariant: This means that the inversion is optimized for a


limited time window.

 Space invariant: This assumes that the data has been processed
optimally to remove spatial variations in the wavelet.

February, 2011 131


There are two basic procedures for wavelet extraction in STRATA:

(1) Use the well(s) and


seismic to extract both
the amplitude and phase
spectra of the wavelet.

(2) Use the seismic alone to


extract the amplitude
spectrum of the wavelet.
Assume the wavelet is
zero phase.

February, 2011 132


Why not always use
the wells? Extract

Log correlation
errors (stretches)
can cause very big
phase problems.

Solution: do log
correlation before
wavelet extraction
using wells.
Extract

February, 2011 133


Statistical Wavelet
Extraction (don’t use
wells):

This procedure uses only the autocorrelation from the seismic data. The
phase is assumed known.

Main parameters:
• Trace range (usually set this large to increase statistics)
• Time window (should be at least twice the wavelet length)
• Wavelet length
February, 2011 134
Example of Statistical Wavelet extraction: Note that the wavelet is zero-
phase because the user has set that as a parameter.

February, 2011 135


Wavelet extraction using well logs:

February, 2011 136


Wavelet extraction using well logs:

This procedure uses the well logs to


estimate both the amplitude spectrum
and the phase spectrum of the
wavelet. It depends critically on the
quality of the tie between logs and
seismic.

Main parameters:
 Select wells to use (use only logs
which tie well)
 Time window
 Wavelet length
 Extraction Type

February, 2011 137


Full Wavelet Option:
This extracts both the amplitude and phase spectrum exactly by solving for
the time-domain operator which shapes the well log reflectivity to the
seismic composite trace. This will only work if the tie is extremely good.

February, 2011 138


Constant Phase Option:
This option calculates the amplitude spectrum of the wavelet using the
autocorrelation of the seismic traces, exactly as in the statistical procedure.
The phase spectrum is approximated as a single constant value, using the
well logs. This procedure is more robust than the Full Wavelet calculation,
especially when the tie between logs and seismic is poor. This is the default
choice.

February, 2011 139


Steps for calculating the phase:

(1) Calculate the wavelet using the statistical wavelet extraction


procedure (don’t use the wells).
(2) Apply a series of constant phase rotations to the extracted
wavelet.
(3) For each phase rotation, calculate the synthetic trace and
correlate it with the seismic trace.
(4) Select the phase rotation which produces the maximum
correlation.

February, 2011 140


If the well tie is good, the methods
produce similar results:

Constant Phase Full Wavelet

February, 2011 141


A General Problem with wavelet extraction:

 To extract a wavelet using logs, an optimum correlation must be


done first.
 To perform correlation properly, the wavelet must already be
known.

Practical wavelet extraction procedure:

(1) Use statistical wavelet extraction to determine a preliminary


wavelet. This assumes that the approximate phase of the
wavelet is known.
(2) Stretch/squeeze the logs to tie the seismic data.
(3) Extract a new wavelet using the well logs.
(4) Possibly repeat steps (2) and (3).

February, 2011 142


A very useful wavelet diagnostic is the Cross Correlation window.

Maximum correlation after the


current well is shifted.

Suggested shift of the well.


This is only (exactly) correct if
no stretching is required.

The symmetry of these side


lobes shows that there is no
residual phase error.

February, 2011 143


If there is more than one well, a very good diagnostic
is Multi-well Analysis:

February, 2011 144


The Correlation Plot shows a graph of correlation
for each well. This can be used to flag bad wells,
which can be removed from a later wavelet
extraction.

February, 2011 145


Exercise 3: Blackfoot – Starting the Project

In this exercise, we will begin inverting a new data set. We will use the new
HRS9 version of the Hampson-Russell software suite in this exercise.

This data set is from the Blackfoot area of Western Canada, and consists of
13 wells which tie a 3D volume. These wells have already been loaded into a
GEOVIEW database. The first step is to start the HRS9 Geoview program.

Start the HRS9 Geoview program by


clicking HRS9 Geoview icon on your
desktop:

February, 2011 146


When the Geoview
program appears, it shows
the Start Page, which
contains a list of
previously opened
projects. Your list may be
empty.

Click on Create New


Project:

February, 2011 147


Type in the project name “blackfoot” and
click Ok:

February, 2011 148


By default, Geoview
expects to use a well log
database with the same
name as the project,
located in the same
directory. If that is not the
case, you can Specify the
database.

In this case, we have


created the database
previously, with the 13
wells already loaded. So,
click Specify database and
Open.

February, 2011 149


On the File Selection
dialog, select the file
blackfoot.wdb and click
Ok:

Finally, click Ok on the


Specify Database menu to
create the new project:

February, 2011 150


The Geoview Start Window now looks like this:

February, 2011 151


Displaying the wells

One part of the Geoview window (called


the Project Manager) shows all the
project data so far.

The tabs along the left side select the


type of project data. Right now, the
Well tab is selected and we can see the
13 wells from the external data base.

Click the “+” sign near one of the wells


(01-17 is shown as an example), to see a
list of curves in that well:

February, 2011 152


To see more details about the wells,
click the Data Explorer tab to the right:

The Geoview
window now
changes as
shown:

February, 2011 153


Click the arrow next to any of
the wells (for example, well 01-
17) to get more information
about the curves in that well:

Click this to return to the


previous table:

February, 2011 154


To the right of the workspace, we can
see a base map, showing the location
of the wells:

Below the base map are a series of


tabs:

February, 2011 155


Click the Single Well Display tab:

This shows the curves for the


selected well:

February, 2011 156


Finally, to see the most complete view
of the log curves within a well, go to the
icon for that well within the Project Data
window and double-click. In this case,
we will choose well 01-08:

This creates a new tab within


the main Geoview window,
called the Wells tab, which
displays the selected well
curves:

February, 2011 157


Loading the Seismic Data

We have now loaded the wells which will be


used in the post stack inversion process. The
next step is to load the seismic volume.

On the far left side of the Geoview window


and click the Seismic tab:

The window to the right of this tab


shows all seismic data loaded so far.
This is empty. Go to the bottom of the
window and click the Import Seismic
button:

On the pull-down menu, select From


SEG-Y File:

February, 2011 158


On the dialog that appears,
select the file
blackfoot_seismic.sgy and
click Next:

Set the Geometry Type to


3D and click Next:

February, 2011 159


On the third page, we are telling the
program what information it can use
from the trace headers. In fact, in
this data set, there are Inline and
Xline numbers, but not X and Y
coordinates. That is why we answer
No to the question “Do you have X &
Y coordinates in the trace
headers?”:

After modifying that item, as


shown, click Next to see
the SEG-Y Format page:

February, 2011 160


By default, this page assumes that the seismic data is a SEG-Y file with all
header values filled in as per the standard SEG-Y convention. For example,
it expects to find the Inline and Xline numbers at the byte locations shown
above. If you are not sure that is true, click Header Editor to see what is in
the trace headers.

In our case, we believe the format information is


correct, so click Next to move to the next page.
Now the following warning message appears
because the program is about to scan the entire
SEG-Y file.

Click Yes to begin the scanning process.

February, 2011 161


When the scanning has finished, the
Geometry Grid page appears:

Because we have read the Inline and


Xline numbers from the SEG-Y headers,
the geometry is correct. Click OK. After
building the geometry files, a new
window appears, showing how each of
the wells is mapped into this seismic
volume:

Click OK to accept the locations shown


on this window.

February, 2011 162


Now the seismic data appears within the Geoview window:

February, 2011 163


Modifying the Seismic Display

The Geoview window currently shows Inline 1.


We will now look at other parts of the data.
The first thing to see is the Base Map. To do
that, select View>Base Map:

The base map appears, showing that


the seismic volume is a rectangular grid
with 119 inlines and 81 cross lines. You
can also see the 13 wells located within
the volume:

After looking at the base map, close


that window by clicking on the “x”
on the upper right corner of the
map.

February, 2011 164


Now let’s look at a different inline. Type the
number 27 as shown, and press the Enter key:

Inline 27 now appears. At


the same time, we can see
one of the sonic logs.
Scroll down to see this
view:

February, 2011 165


To see one of the cross lines, click the field
which currently says Inline. Select the Xline
option.

Cross line 1 now appears. To see the


display positioned at one of the well
locations, go to the Well icon and click
the down arrow as shown:

The drop-down menu shows a list of


wells in the project. Select one – say,
08-08, as shown – and the Geoview
window shows the crossline which
intersects that well location.

February, 2011 166


We can also modify other plotting
parameters by using the Seismic View
Parameters window. To bring up that
window, click the “eyeball” icon as shown:

The Seismic View Parameters window


contains a series of pages which
control various aspects of the plotting.

To see the parameters for a specific


item, select that item from the list at the
left side. For example, here we have
selected the Inserted Wells item:

February, 2011 167


Let us (temporarily) insert the
density log by selecting that item
as shown:

Now click Apply on the Seismic


View Parameters window. The
display is modified accordingly:

February, 2011 168


We will continue the
exercise with the sonic logs
reinserted. To do this, click
Reset Page and OK on the
View Parameters window.

This redraws the Geoview


window as before.

February, 2011 169


Importing Horizons
The last data component required for building the initial inversion model
is a set of horizon picks. You can use Geoview to pick the data directly.
Alternatively, you can import horizons which have been previously
picked in other software.

To start that process, select Horizon >


Import Horizons > From File:

From the File Selection


Window, highlight the file
called
blackfoot_horizons.txt
and click Select.

Note that, at the lower


left corner of the dialog,
we are specifying this to
be a Free Format file.
Click Next:
February, 2011 170
The next page of the dialog allows you
to specify how the file is organized:

Click the View Files button to see the


ASCII file:

February, 2011 171


The file display shows that there are 2
horizons in the file, and that we need to
skip the first 4 information lines.

Fill in the format dialog


as shown, including the
new Horizon Names:

February, 2011 172


When you have
modified the
dialog, click OK
and the imported
horizons will be
displayed on the
seismic window:

February, 2011 173


Using the Post Stack Inversion Workflow

Now that we have read in all the data necessary for the inversion, we are
ready to start the process.

First, look at the horizontal tabs to the


left of the seismic window. You will see
that one of those tabs is called
Processes. Click that tab to see a list of
all the operations which are available in
Geoview. Each of the processes can be
expanded. For example, if you click on
both the Seismic Processing and
Inversion options, the following
expanded list is seen.

One way to do the inversion would be to


apply each of the desired options in
turn.

February, 2011 174


We will use an alternate procedure in this
tutorial. We will use the pre-defined
Workflows. Click the Workflows tab. The
window changes like this:

Each of the items on this window contains


a complete workflow for the specified
process. Click the item called Post Stack
Inversion. The window changes like this:

We now see the suggested series of steps


to be followed for Post Stack Inversion. The
steps are colored red to indicate that the
parameters have not yet been supplied.
These are the “default” steps, but the list
can be edited and customized, as we will
see later.
February, 2011 175
Double-click the first item Select Post-
stack Seismic. An arrow will appear in
front of the item, as shown here:

Now a dialog appears on the right


with a list of all seismic volumes in
the project:

Since we have only loaded one seismic


volume, that volume is selected. Note
that at the lower right corner of the
dialog, there is a button for importing
more seismic volumes:

In fact, we want to use the selected seismic volume,


which is highlighted, so click Select on this dialog.
February, 2011 176
Now double-click the second item of the
workflow, Select Horizons:

The dialog on the right shows the two


horizons we have just loaded:

It also contains buttons for picking or


importing new horizons:

Click Select to accept the two horizons.

February, 2011 177


The next step is to apply the process
Extract Statistical Wavelet, by double-
clicking that option.

There are two basic methods for extracting the wavelet. One method uses
the wells, and can give a good estimate of both amplitude and phase spectra
of the wavelet.

The second method – called “statistical” – uses the seismic data alone to
extract the wavelet. This method will estimate the amplitude spectrum from
the seismic data, but we must make an assumption about the phase –
typically we assume the data are zero phase.

In this step, we are extracting a statistical wavelet. We will refine the wavelet
extraction using the wells at a later stage.
February, 2011 178
The dialog on the right sets the
range of data to analyze:

By default, the program will


examine the entire data volume,
but this is rarely appropriate. In
particular, we want to set a time
window around the zone of
interest. Change the dialog to
extract using the limited time
window shown on the single
cross line now displayed on the
screen:

February, 2011 179


When you have changed the dialog as
shown above, click Run to extract the
wavelet. The extracted wavelet appears
in its own pop-up window:

Note that the time domain response is


in the upper window, while the
amplitude and phase spectra are in the
lower.
Note also this small button at the lower
right of the wavelet window:

February, 2011 180


If you click that button,
the wavelet window will
be “docked” within its
own Wavelets tab:

This is a handy way to keep track of any window created within


our software. To release the wavelet window from its tab, click
on the “airplane” at the lower right of the wavelet window:

All the windows created within our software can be docked or


floating in this way. Finally, send the wavelet window back to
the wavelets tab by clicking the Wavelets button once again:
February, 2011 181
The next step is to select the wells which
will be used in the model building:

Once again, we see a list of the wells


which have already been loaded into the
project:

Click Select to complete this step.

February, 2011 182


The Wells tab appears,
showing one of the wells in
the project:

You can apply various log processing


options, like Log Editing, by going back
to the Processes list:

For this exercise, we will assume that


the logs have been properly edited.
Return to the Post Stack Inversion
Workflow.
February, 2011 183
Correlating the Wells

The next step in the workflow is to


Correlate the Wells, so double-click that
item:

In practice, each of the wells used to build


the inversion model needs to be
correlated. For this tutorial, we will
correlate just one of the available wells,
and assume that all the others have been
correlated previously. On the Well
Selection Dialog, select the well 08-08 and
click OK at the bottom of the dialog:

February, 2011 184


Now a dialog appears to specify
which seismic volume will be
used for the correlation process,
and how the composite trace will
be extracted from that volume:

Click OK on this dialog.

February, 2011 185


The Log Correlation
Window now appears:

The blue traces on this


display are synthetic
traces calculated from the
sonic and density logs in
this well, using the depth-
time curve currently
stored in the database
and the wavelet we have
previously extracted:

February, 2011 186


The plot at the upper right shows the cross
correlation between the synthetic trace and
the composite trace:

That correlation result depends on the analysis


window, which can be improved. The cross
correlation window defaults to be the largest
possible window containing both the synthetic
and real trace. This can usually be improved
by narrowing the analysis to the region when
the log tie is best:

Set the start time to 800, as shown above


and click on Apply.
February, 2011 187
The correlation plot now shows a maximum
correlation of 71%.

It also suggests that the synthetic


should be shifted down by 6 ms.

That information is also displayed on


the menu bar at the base of the window:

Click the Apply Shift button to apply the


suggested 6 ms shift.

February, 2011 188


The asymmetric shape of the Cross
Correlation plot suggests that there is a
residual phase error in the synthetic,
which could be improved by extracting a
new wavelet now using the wells. To do
this, click the Wavelet button and choose
the option Extract Wavelet Using Wells:

On the dialog, change the parameters


as shown, and click Run:

February, 2011 189


The extracted wavelet
shows an average phase of
-53 degrees:

Click the Wavelets button


to move this new plot to
the Wavelets tab.

February, 2011 190


The Correlation Plot now shows a maximum
correlation of 74%, with a more symmetrical
shape. From this we can conclude that we have
a good estimate of the wavelet phase.

The plot also suggests a further shift of


-2ms. Click Apply Shift.

Click OK to accept this correlation.

A dialog appears suggesting a name for


the new sonic log we have created by
the log correlation process. Click OK to
accept that new name:
February, 2011 191
We have now completed the log correlation step for one of the 13
wells, and extracted a single wavelet, for all the wells.
In a real project, we would have to do this correlation step for the
remaining 12 wells. To save time, we have done this correlation for
you, so we will assume the other wells are correlated.
The complete flow for a general multi-well project is:

(1) Extract a single Statistical Wavelet.


(2) Go through each of the wells, doing the correlation.
(3) Extract a single wavelet using all the wells.
(4) Go through each of the wells again, fine tuning the correlation.
Usually this step means simply apply a bulk shift.

(End of Exercise 3)
February, 2011 192
Model Based Inversion Parameters

The parameters for model based inversion:

The most significant


parameters are:

• Number of Iterations
• Average Block Size
• Type of scaling

Less important parameters


are:

• Inversion Option
• Maximum Impedance
Change

February, 2011 193


Number of Iterations

Since STRATA converges through a series of iterations, this parameter


determines the degree of convergence. In practice most of the work has
been done after about 3 iterations.

There is never any harm in having more iterations - it only affects the run-
time.

The number of iterations required for convergence may depend on the


block size used in the inversion. A finer block size may require more
iterations.

The way to confirm whether enough iterations have been done is to


examine the error plot.

Recommendation : Use 10 or more iterations.


February, 2011 194
Average Block Size

This parameter controls the resolution of the final result. The initial guess
model is blocked to a series of uniform blocks with this size:

The final inversion result may change the size of the blocks, but the number
of blocks is still the same. This means that some blocks get bigger and
some get smaller, while the average is kept constant.

Using a small block size (2 ms) will increase the resolution, but the increased
detail may be coming from the initial guess. This may be alleviated by using
a smooth initial model. Using a small block size will always improve the fit
between the input trace and the final synthetic trace.

February, 2011 195


Average Block Size

Inversion using 6
ms block size:

Inversion using 2
ms block size:

Recommendation: Set block size equal to sample rate, but use smooth initial
model.
February, 2011 196
Scaling Parameters

In addition to the main Post-stack inversion Parameters, the following


page controls the scaling of the data:

February, 2011 197


Why is Scaling an Issue?

The Convolutional Model is used as the basis for all inversion:

Trace = Wavelet * Reflectivity + Noise

In the frequency domain, this can be approximated by:

Reflectivity = Trace / Wavelet

To solve for the reflectivity, the wavelet must be known.

This means that the relative amplitudes of the reflection coefficients depends
on the absolute scaling of both the Trace and Wavelet.

From the equation above, if the wavelet is multiplied by 2, the resulting


reflectivity will be divided by 2.

STRATA determines the scaling of the trace automatically by forcing the root-
mean-square amplitude of the initial guess synthetic to be equal to the root-
mean-square amplitude of the real trace.

February, 2011 198


Scaling Options

The second option, single global scaler, is theoretically more desirable. This
is because it assumes that there is a single wavelet scaling which is suitable
for all traces of the data set. This will preserve amplitude variations from trace
to trace.

The first option, separate scalers, is can be more robust for noisy data. It
effectively assumes that traces may need to be rescaled to remove trace-to-
trace variation which is not based on lithology.

For some data sets, especially sparse models, the automatic scaling may not
be ideal. In that case, you may override with a manual adjustment, which
multiplies the automatic scaling result:

The only way to determine this factor is by visually inspecting how well
the inversion traces match the initial guess logs.
February, 2011 199
Scaling Scaling Scaling
Scaling Comparison
too low too high just right

Blue = original
impedance log

Red = inversion
impedance
result

February, 2011 200


Scaling Options

Recommendation: Use inversion analysis to calculate a


single global scaler at the well locations.

February, 2011 201


Inversion Option

This parameter controls the how the constraints will be used.

Model Based inversion minimizes an objective function of this form:

J = weight1 x (T - W*r) + weight2 x (M - H*r)

where:
T =
the seismic trace
W =
the wavelet
r =
the final reflectivity
M =
the initial guess model impedance
H =
the integration operator which convolves with the final
reflectivity to produce the final impedance
* = convolution
February, 2011 202
The objective function has two parts.

Minimizing the first part, (T - W*r), forces a solution which models the
seismic trace. Minimizing the second part, (M - H*r), forces a solution which
models the initial guess impedance using the specified block size.

These two conditions are (usually) incompatible. The weights, weight1 and
weight2, determine how the two parts are balanced. In Soft Constraint
inversion, the objective function is exactly as shown above. The weights are
determined by this parameter:

The Model Constraint is the value of weight2 in the objective function. Setting
this value to 0 causes the seismic trace to dominate. Setting this value to 1
causes the initial guess model to dominate. This is called a soft constraint
because the final model may deviate any distance from the initial guess, but it
pays an increasingly large penalty for doing so.
February, 2011 203
In Hard Constraint inversion, the second term is missing entirely from the
objective function. However, the algorithm is constrained to keep the final
impedance values constrained within the limits specified by:

This is called a hard constraint,


because values are not allowed to
change beyond a fixed boundary.

The Maximum Impedance Change is a percentage of the average


impedance for the log. Note the effective range for this model:

Recommendation: Use Hard Constraint with default parameters.


February, 2011 204
Error Plot

The Error Plot shows the difference between the actual traces and the
synthetic traces calculated using the inversion impedance result:

Ideally, the
Error Plot
should
show no
coherent
energy, and
should
have a low
over-all
amplitude.

February, 2011 205


Low frequency component in the error – probably caused by using the
wrong wavelet:

February, 2011 206


Error localized to one side of line – probably caused by not picking enough
events:

February, 2011 207


Coherent error throughout data set – probably caused by:
 too large block size
 not enough iterations
 constraint too tight

February, 2011 208


Exercise 4: Blackfoot – Model-Based Inversion

We are now ready to perform the next step on


the workflow, which is Build Initial Model.

Double-click that item on the workflow list:

The dialog which appears


contains the default parameters
for building the standard post-
stack inversion model. By
default, all the wells are selected:

We will use both horizons in the


project:

February, 2011 209


After interpolating the well log curves, a low-pass frequency filter
will be applied, which, by default, passes all frequencies up to 10
Hz, filters all frequencies above 15 Hz, and interpolates the filter
between those limits.

There are many


additional “Advanced”
parameters, which you
can examine by clicking
the Show Advanced
Options button:

For this tutorial, click Run to accept the


defaults.

February, 2011 210


The Geoview window
now shows the
calculated initial
model within the
Models tab:

The wiggle traces on


this display are the
original seismic data,
while the color
displays the filtered
acoustic impedance.
You can display any
location on the model
volume by using the
selection tools on the
menu bar:

February, 2011 211


Performing Inversion Analysis

The next step is to perform Inversion


Analysis. This is the process of running
inversion at the well locations to QC and
optimize the inversion parameters. At the
same time, scalars are automatically
determined which scale the input seismic
data to the amplitude range of the
synthetic seismic data. On the workflow,
double-click Inversion Analysis:

The dialog which appears contains default


selections of the main parameters. These are
usually appropriate.

In this case, we are using the seismic data


called blackfoot_seismic and we are inverting
the entire time window:

February, 2011 212


We are using all the wells in the
inversion analysis. Note that you
can use different wells than the
wells used to build the initial model:

We are using the initial model


previously generated:

We are using the wavelet previously


extracted:

To verify that is the right wavelet, click


Change Wavelets > Wavelet Data
Explorer, as shown.

February, 2011 213


A dialog appears, showing
you the details of the
wavelet:

Click the “x” on the upper


right corner to dismiss
this dialog.

Now click OK on the Inversion Analysis


dialog to start the process:

February, 2011 214


The Inversion Analysis window now
appears, showing the inversion result at
the first well location:

From left to right, the display shows


the inversion result (in red) overlaying
the original impedance log from the
well. To the right of that, we see the
synthetic traces calculated from this
inversion result (in red) followed by the
original seismic composite trace (in
black). Finally, we see the error trace,
which is the difference between the two
previous results.

We are seeing the result at the first well


location (01-08), but the controls on the
upper menu bar allow us to see any
other well:
February, 2011 215
To the upper right of the display are a
series of buttons, which allow us to fine-
tune the inversion. For example, the
Wavelet button allows you to manipulate
the inversion wavelet.

A second button allows us to view and


change the initial model parameters:

A third button brings up a dialog


allowing us to change the inversion
parameters. Click this button as shown.

February, 2011 216


For example, change the Inversion Method
to Linear Programming Sparse Spike and
click Apply at the bottom of the dialog.

The inverted traces now show a


blockier appearance:

February, 2011 217


Set the Inversion Method back to Model
Based and click Apply to restore the
default settings:

Then close the Inversion Parameters


dialog by clicking the Close button:

February, 2011 218


Applying the Inversion

Now that we have optimized the


parameters at the well locations, the
last step is to apply the inversion to the
entire 3-D volume. Double-click the last
item on the workflow:

The dialog which appears shows all the


inversion parameters, but we do not have
to change any, because we have already
optimized them at the well locations.

The only significant parameters involve


the data range and time range of the input
volume to be inverted:

In this case, we will invert the entire


volume, so click Run at the base of the
dialog to start that process.
February, 2011 219
When the inversion process is done, the result is displayed
in a split-screen along with the initial model:

February, 2011 220


Displaying the Inversion

The default split-screen display is very useful for looking at the results, but
there are many improvements possible.

For example, you can increase the


available plot space by clicking on
the “x” on the Project Manager
window, as shown, to temporarily
hide that window:

To restore the Project


Manager window, click its
name to the left:

February, 2011 221


You can also temporarily hide one of
the views. For example, click on the
first icon shown below to temporarily
hide View 1, which shows the model:

To restore View 1, click it again:

There are actually 3 views


available. Click on the third icon
to display View 3:

February, 2011 222


The third view is currently blank:

To load some data into View


3, first go to the Project Data
window on the left and find
the input seismic data:

Then, holding the left mouse


button down, drag-and-drop
the volume blackfoot_seismic
into the blank View 3:

February, 2011 223


The window now looks like this:

The fourth button sets the


orientation horizontally:

February, 2011 224


Now turn off Views 1 and 3 by clicking
the first and third buttons:

Now, right-click within the inversion


window. A series of display options
appear for this window. For example,
we can easily modify the Color Scheme.

February, 2011 225


The Color Scale numerical range can be
changed graphically by selecting Color
Key > Color Key and Histogram:

Click Cancel to remove this


display:

February, 2011 226


Set the range manually by selecting
Color Key > Modify Range:

On the dialog which appears, set


the desired range from 8000 to
12000 and click OK:

February, 2011 227


Another useful display is a data slice
through the inversion volume. To
produce that, go to the tab called
Processes. From the list select double-
click Create data slice:

February, 2011 228


Fill in the menu as shown, and click Ok to
produce the slice:

(End of Exercise 4)
February, 2011 229
Maximum-Likelihood Sparse Spike
Inversion Parameters

The menu for sparse spike inversion:

Sparse Spike Inversion uses the same parameters as constrained model


based inversion.

These additional parameters determine how many spikes will be detected


on each trace:

Maximum Number of Spikes


Spike Detection Threshold

February, 2011 230


Maximum Number of Spikes

This parameter sets the maximum number of allowable spikes per trace.
This is defaulted to be the same as the total number of samples in the
window. Effectively this means that this parameter does not operate under
normal conditions.

Spike Detection Threshold

As each spike is added, its amplitude is compared with the average


amplitude of all spikes detected so far. When the new amplitude is less
than a specified fraction of the average, the algorithm stops adding spikes.

February, 2011 231


Linear-Programming Sparse Spike Inversion
Parameters

LP sparse spike inversion minimizes an objective function of this form.

J  weight1 * T  W * r   weight 2 *  ri

The first term tries to produce an impedance result whose synthetic matches
the input seismic trace. The second term is a constraint which favors
solutions with “sparse” reflectivity or “blocky” impedances”.

The LP Inversion
parameters are
shown here:

February, 2011 232


Sparseness

The most important parameter is the Sparseness, which controls the


relative weighting of the two terms:

1% 50% 100%

February, 2011 233


Window length

The LP Inversion algorithm is very time consuming. To decrease run-time,


the inversion is run over a series of small overlapping windows.
Theoretically, a larger window is always preferable, at the cost of
increased run-time:

February, 2011 234


Output Impedance Type:

This parameter outputs either the abolute impedance using the low-
frequency model (Full Spectrum) or the relative impedance without the
low-frequency model (High Frequency Residual)

February, 2011 235


Bandlimited Inversion Parameters

The menu for bandlimited inversion:

The only parameter for Bandlimited Inversion is:

Constraint High-Cut Frequency: This parameter controls the filter which is


applied to the initial guess model to provide the low-frequency component
to the result. All frequencies above this value are removed from the initial
guess. All frequencies below this value are removed from the recursively
inverted trace. The two are then added together.

February, 2011 236


Colored Inversion Parameter Menu

First, you must run


spectral analysis and
create an operator. The
result is shown on the
next slide.

One key parameter for


Colored Inversion is the
Impedance Output
Option, where High
Frequency Residual, or
Relative impedance is
the default.

The other option is to


create a Full Spectrum
by the adding the
specified frequency
range2011
February, from the model. 237
Colored Inversion Operator
Calculation

The spectral analysis and operator creation result has two parts. The top
part of the display shows the analysis and operator results.
February, 2011 238
The operator is created automatically using a least-squares fit. The bottom
part of the display allows you to change the automatic parameters if you
wish, and is shown above.

The next set of slides describes the parameters on this menu.

February, 2011 239


The first panel shows a plot of
the amplitude spectrum of
Impedance from a series of wells
vs Frequency. This is shown on
a log/log scale.

The red line is a regression


curve, which represents the
“desired output” of the Colored
Inversion.

These parameters allow you to


over-ride the automatic calculation
of the regression line:

February, 2011 240


The second panel shows two
curves. One is the amplitude Desired Spectrum
spectrum of the input seismic
data. The second is the “desired
Seismic Spectrum
output” from the previous panel.
Note that this is now curved,
because we are showing a linear
scale in Frequency.

This parameter allows you to apply


smoothing to the Seismic
Spectrum:

February, 2011 241


The horizontal red line is the
Spectrum Threshold. This sets a
frequency range over which the
inversion operator will be
calculated. Only those frequencies Spectrum Threshold
for which the seismic spectrum
(blue) is above the threshold will be
used in the calculation. The
threshold prevents division by zero
or small noise values.

This parameter allows you to


change the threshold value:

Alternatively, these parameters (if set


to non-zero values) allow you to set
the frequency range manually:

February, 2011 242


Finally, these panels show the time and frequency domain operator which
has been calculated.

These operator displays will be updated automatically to reflect changes to


all the other parameters on the menu.

These parameters directly affect the


operator itself:
February, 2011 243
Exercise 5: Blackfoot – Other
Inversion Methods

Since we have built the initial model for the Blackfoot data set, we can
easily apply other inversion methods to the data and compare the results
with model-based inversion.

We will start with Bandlimited inversion.

Go to the Process tab and double-click on


Inversion / Process / Post-stack Inversion

February, 2011 244


Fill in the menu as shown. Note that we
choosing to apply the Inversion Method
Bandlimited to a single cross line (42).
When you have filled in the menu, click Ok
to run the process.

February, 2011 245


When the process has finished, it looks like this:

Notice that the


bandlimited
inversion runs
much faster than
model-based
inversion.

However, there is
less detail in the
result.

February, 2011 246


Next we will run sparse spike inversion.

Go to the Process tab again and double-


click on Inversion / Process / Post-stack
Inversion

February, 2011 247


Fill in the menu as shown. Note that we
choosing to apply the Inversion Method
Linear Programming Sparse Spike to a
single cross line (42). When you have filled
in the menu, click Ok to run the process.

February, 2011 248


When the process has finished, it looks like this:

Note that the


detail is
comparable to
model-based
inversion, but
it is not as
continuous
laterally.

February, 2011 249


Finally we will run sparse colored
inversion.

Go to the Process tab again and double-


click on Inversion / Process / Post-stack
Inversion

February, 2011 250


Change the menu as shown.

Before running the process, we have


to create the Colored Inversion
operator. To do that, click the button
Run spectral analysis and create
inversion operator:

February, 2011 251


Click Ok on the two menus which
follow:

February, 2011 252


Finally click Ok on the main inversion
menu:

February, 2011 253


When the process has finished, it looks like this:

February, 2011 254


We have now produced a series of
inversion volumes. To see a list of
the volumes, click on the Project tab
and expand the entries by clicking
the “+” signs as shown:

We can see four inversions. Each


inversion contains the inversion
result (with the suffix “_Zp) along
with accompanying synthetic and
error volumes.

February, 2011 255


We will arrange several of these results in a single window for comparison.

First the model-based inversion. Double click on “inverted_Zp” as shown:

This causes
the model-
based
inversion to
appear.

February, 2011 256


Now create two
empty windows
by clicking the
“2” and “3”
icons on the
lower right:

February, 2011 257


Select the second volume “lp_inversion_Zp” and drag it to the empty View 2:

Similarly, drag the third volume “colored_inversion_Zp” to the empty View 3:

February, 2011 258


Finally, click the fourth button on the lower right to switch between vertical
and horizontal views:

February, 2011 (End of Exercise 5) 259


Summary

Inversion is the process of extracting, from the seismic data, the


underlying geology which gave rise to that seismic.

Inversion can be a very non-unique process. The low-frequency model


is particularly important.

Successful inversions depend on careful correlation of each of the wells


and careful wavelet extraction.

In this course we have studied the following types of inversion:

Model-based inversion
Bandlimited inversion
Colored inversion
Sparse-spike inversion

Model-based inversion is the recommended choice for most inversion projects.


February, 2011 260
Appendix: Pre-stack, or AVO, Inversion

The basic convolutional model assumes zero-offset data.

Conventional inversion should not be applied to data with AVO


effects, since changes in VP/VS are not explicitly accounted for.

To extend inversion to handle AVO data, these algorithms are


currently used:

(1) Elastic Impedance


(2) Independent Zp and Zs inversion
(3) Simultaneous Inversion for Zp, Zs, and density
(4) Lambda-Mu-Rho (LMR)

These techniques will now be discussed, followed by an exercise on


simultaneous inversion and LMR.

February, 2011 261


Elastic Impedance
The Elastic Impedance concept was originally developed by Connolly
(The Leading Edge, 18, no. 4, 438-452 (1999)).

He started with the Aki-Richards equation which relates reflection


amplitude to incidence angle:

 VP  VS 
RP (q )  a b c , where :
2VP 2VS 2
2 2
 VS   VS 
a  1  tan q , b  8  sin q , and c  1  4  sin 2 q .
2
  2

 VP   VP 

Note that post-stack inversion theory 1   VP  


assumes that q = 0, which gives us: RP (0 )  
o
 
2  VP  
Thus, changes in VP/VS are ignored.
February, 2011 262
Elastic Impedance
Notice that, for the zero-offset case:

1   VP   1 AI 1
RP (0 )  
o
    ln ( AI )
2  VP   2 AI 2
where AI  Acoustic Impedance
By analogy, Connolly defined a new type of impedance such that:

1 EI 1
RP (q )   ln ( EI ) , where EI  Elastic Impedance.
2 EI 2
By mathematical manipulation, he showed that:

(1 tan 2 q ) ( 8 K sin2 q ) (1 4 K sin2 q )


EI (q )  V P V S  V V  ,
a b
P S
c

2
V 
where K   S  .
 VP 
February, 2011 263
Elastic Impedance

This figure, from Connolly’s paper shows an overlay of Elastic Impedance


over Acoustic Impedance from a well. The Elastic Impedance shows
anomalously low values at hydro-carbon areas.

February, 2011 264


Colony sand case study

 In the following set of slides, we will consider a


case study from the Colony sand in Alberta.
 This is a 2D example which lends itself well to AVO
analysis.
 The analysis was done using the Hampson-Russell
AVO program.
 Note the dramatic change in the elastic impedance
response when we invert for Elastic Impedance at
two different angles.

February, 2011 265


Gas sand case study

The figure above shows the logs after fluid substitution in the gas zone. The EI_Near
log on in blue was created at 7.5o and the EI_Far log in red was created at 22.5o. Note
that the Near < Far outside the gas sand but Far > Near inside the sand.
February, 2011 266
Gas sand case study
EI_Near EI_Far

(a) (b)
The figure above shows (a) the interpreted crossplot between the near and
far EI logs, and (b) the zones marked on the logs themselves. Notice the
clear indication of the gas sand zone.
February, 2011 267
Elastic Impedance

The work flow for this type of inversion starts from the pre-stack data,
creates two angle stacks, and inverts each separately.

Gathers

AVO Analysis
AVO
Program Near angle Far angle
stack at q1 stack at q2
STRATA
Program Invert to EI(q1) Invert to EI(q2)
February, 2011 268
Elastic Impedance

Far Angle Inversion (22.5o)

This produces 2
inversion results.

Note the improved


definition of the
gas sand on the
far angle
inversion Near Angle Inversion (7.5o)

February, 2011 269


Gas sand case study

The figure on the left shows a crossplot


between the EI at 7.5o, on the horizontal axis,
and the EI at 22.5o, on the vertical axis. The
background trend is the grey ellipse, and the
anomaly is the yellow ellipse. As shown
below, the yellow zone corresponds to the
known gas sand.

February, 2011 270


Independent inversion for P and
S-Impedance
 We now turn from elastic impedance inversion to independent P and S-
impedance inversion. Later, we will discuss simultaneous inversion
and how it differs from independent inversion.

 Both independent and simultaneous inversion for P and S-impedance


will lead us to the lambda-mu-rho (LMR) approach, but we will talk
about LMR before introducing the simultaneous approach.

 We call this method “independent” inversion, because the first step is


to extract independent estimates of the zero-offset P and S
reflectivities, RP0 and RS0 from the seismic gathers. This is done using
the Fatti equation:

RPP (q )  c1RP 0  c2 RS 0  c3RD


2 2
V  1 V 
where c1  1  tan q , c2  8 S  sin 2 q , c3  tan 2 q  2 S  sin 2 q ,
2

 VP  2  VP 
1   VP   1   VS   
RP 0      

   
, R , and R .

S0 D
2  VP 2  VS
February, 2011 271
RP and RS Inversion Flow

Gathers

AVO
AVO Analysis
Program
RP Estimate RS Estimate
STRATA
Program Invert to ZP Invert to ZS

A flow chart for the independent inversion procedure. Note that


both the AVO and STRATA programs are required.
February, 2011 272
RP and RS Sections

Here are the RP


and RS sections,
extracted using the
AVO program, with
the correlated P-
wave sonic
inserted at the
proper location,
and three picked
horizons. Horizon
2 is picked on the
gas sand trough.

February, 2011 273


P-wave and S-wave Models
Here are the initial P-impedance Model
models for inversion.
Note that these
models were created
under the Model/
Build/Rebuild a Model
option using:

S-impedance Model

February, 2011 274


P-wave and S-wave Inversions

Here is the final P-wave


and S-wave inversion
results. The low P-wave
impedance just below
Horizon 2 represents the
gas sand.

Note that this


corresponds to an
increase in S-wave
impedance.

February, 2011 275


Lambda-Mu-Rho (LMR)

The Lambda-Mu-Rho or LMR method was originally proposed by


Goodway et al (SEG Expanded Abstracts, 1997). Like the Elastic
Impedance method, this procedure extends conventional inversion to
handle data with AVO effects.

LMR uses the following relationships between VP, VS,  and the Lamé
parameters, l and m:

l  2m m
VP  and VS  Note that the final result
  is to express the
quantities l and m in
therefore : m  Z 2
S
terms of the acoustic
impedance ZP and
and : l  Z  2 Z
2
P
2
S
shear impedance ZS.

February, 2011 276


LMR Flowchart

The work flow for LMR involves Gathers


calculating RP and RS seismic
volumes from pre-stack data.

Two inversions are performed


AVO Analysis
to create ZP and ZS volumes.

These volumes are transformed RP Estimate RS Estimate


and cross-plotted using the
equations from Goodway et al.
Invert to ZP Invert to ZS

Transform to l and m
Cross-plot
February, 2011 277
LMR Example
l
This example shows the
result of applying the
LMR approach to a gas
sand example from
Alberta, where the gas
sand is indicated by the
ellipse.

The top section shows m


the lambda-rho result,
and the bottom section
shows the mu-rho
result.

February, 2011 278


LMR Example

This mu-rho vs lambda-rho crossplot is


shown on the left, where the red zone
indicates gas (low lambda-rho) and the
blue zone indicates the shales and wet
m sandstones.

These zones are displayed on the section


below and indicate the gas sand zone.

l

February, 2011 279


Simultaneous Inversion
Simultaneous Inversion inverts for ZP , ZS , and possibly Density using pre-
stack angle gathers as input. The benefit of this procedure is that it allows
constraints to be imposed between these variables. This can stabilize the
results and reduce the non-uniqueness problem.

We again start with Fatti’s version of the Aki-Richards’ equation. This


models reflection amplitude as a function of incident angle:

RPP (q )  c1RP  c2 RS  c3RD


where:
c1  1  tan 2 q , 1   VP  
RP   
2 2  VP  
 VS 
c2  8  sin 2 q , 1   VS  
 VP  RS   
2
2  VS  
1 V  
c3  tan 2 q  2 S  sin 2 q , RD  .
February, 2011
2  VP   280
Simultaneous Inversion

A linear relationship is assumed for the background wet lithologies.


Simultaneous Inversion solves for deviations from this background:

ln( Z S )  k ln( Z P )  kc  LS


ln(  )  m ln( Z P )  mc  LD

Ln(ρ) Ln(Zs)

LD
LS

Ln(Zp) Ln(Zp)
February, 2011 281
Simultaneous Inversion

Simultaneous ZP
Inversion
produces
volumes of Zp, Zs, Zs
Density, and
derived
combinations:
VP/VS

February, 2011 282


Simultaneous Inversion

The interpretation of Simultaneous Inversion volumes is similar to other


AVO Inversion results:

Let us now finish the course with an exercise on Simultaneous Inversion


using a shallow gas sand example from Alberta.
February, 2011 283
Exercise 6: Simultaneous inversion

In this exercise, we will apply pre-stack simultaneous inversion to a single 2D line,


containing a series of angle gathers.

If the STRATA program is still running, close it down by clicking on File / Exit Project
on any of the STRATA windows.

The well for this data set has already


been loaded into a GEOVIEW database.
To access that, click on Database / Open
on the GEOVIEW window:

Select the database


angle_gather_database and click Ok:

February, 2011 284


The database contains a
single well, AVO_WELL.
Highlight that line and click
on Display Well.

The well contains a sonic,


density, and shear wave
log.

February, 2011 285


Now click on the STRATA button to start that
program.

Choose the option to Start a New Project, and


click Ok:

Name the project angle_gather_project and click


Ok:

February, 2011 286


The STRATA window now appears,
initially blank.

The first step is to read in the


angle_gathers, which have already been
created. Click on Data Manager / Import
Data / Open Seismic / From SEGY File:

From the list, select angle_gather.sgy


and click on Next:

February, 2011 287


Select the option to handle this as a
2D Line and click Next>>.

On the next page, specify that this file does NOT


have XY coordinates in the trace headers.

Click Next and Ok until the Well


to Seismic Map Menu appears.
Specify that the well is located
at CDP 330, as shown. Finally,
click Ok on this menu.

February, 2011 288


The angle gather now appears, with the single well inserted.

February, 2011 289


Now, we will import horizons for
model building. Click on
Horizon / Import Horizons / From
File.

Select the file


angle_gather_horizons.txt
and click on Ok.

February, 2011 290


This file has multiple horizons, so click
that option and then click Next>>.

There are 3 horizons, so fill in that


number, and click Next>>

Finally, fill in the format page as


shown to the right. You may want
to Display selected file to verify
these choices. When you have
completed the menu, click on Ok to
load the horizons.

February, 2011 291


Now, we build the initial model for the
inversion. Click on Model / Build /
Rebuild a Model:

On the first page choose the option


Typical setup for Pre-stack Inversion
and click Next>>:

On the next page, we confirm the


wells used in the model. Click on
Next>>.

On the next page, we confirm


which logs curves are used.
Click on Next>>.

February, 2011 292


On the next page, we confirm which
horizons are used in the model
building. Click on Next>>:

Finally, we confirm that the


model will be filtered to retain
only the low frequency
components. Click on Ok.

When the model has been


built, it will look like this:

February, 2011 293


Because this is pre-stack seismic data, the default
display does not show the model in a continuous
form. To verify the model, click on the “eyeball” icon:

On the View Parameters menu, turn off the


trace plotting (temporarily) by changing the
Trace Data Volume to None and clicking Apply:

The STRATA window now shows


the low frequency impedance
model which will be used for the
inversion.

Click on Cancel on the View


Parameters menu to restore the
original STRATA window.

February, 2011 294


For pre-stack inversion, it is usually helpful to
have a wavelet which varies with angle. So far,
in previous projects, we have only extracted a
single wavelet. Now, we will extract two
wavelets, one for the near traces and one for the
far traces.

Click on Wavelet / Extract Wavelet / Statistical:

On the first menu page, set the “Offset”


range from 0 to 15. For angle gathers, this
is actually the angle in degrees.

On the last menu page, set the Wavelet


Name as wave_near.

Click on Ok to get the near angle wavelet.

February, 2011 295


Now, repeat the entire process to get the far
angle wavelet.

Click on Wavelet / Extract Wavelet / Statistical:

On the first menu page, set the “Offset”


range from 15 to 30.

On the last menu page, set the Wavelet


Name as wave_far.

Click on Ok to get the far angle wavelet.

February, 2011 296


Near Wavelet Far Wavelet

By examining the two


wavelets, we can see
a slight loss of high
frequencies for the far
wavelet, as expected:

February, 2011 297


We are now ready to do the simultaneous inversion. This is
actually done in two stages. First we apply inversion at the
well location(s) to confirm the inversion parameters and
allow the program to determine the proper scaling. Then, we
apply inversion to the entire volume. To do the first step,
click on Analysis / Pre-stack Analysis:

On the first menu page, select


angle_gather as the input and click
Next>>:

On the second menu page, we confirm


the angle range for this data set. Click
Next>>:

February, 2011 298


On the next page, we confirm a
number of parameters. The most
important parameter on this page
is the wavelet. By default, STRATA
will use the last extracted wavelet.
To display that wavelet, click on
Set Current Wavelet:

When the wavelet menu appears,


it displays last wavelet we
extracted, wave_far. To use both
the near and far angle wavelets,
click on the option to Set Angle
Dependent Wavelets.

February, 2011 299


Now the menu shows a table,
allowing you to type in the angles
for each of the desired wavelets.

Change the menu to look like this.


Note that we have specified an angle
of 22.5 degrees for the far wavelet.
This is because it was extracted over
a range of 15-30 degrees. Similarly,
the near wavelet is specified as 7.5
degrees, which is the mid-point of the
extraction range 0-15 degrees.

Finally, click on Set Current Wavelet on the bottom left of this menu. The
Analysis Setup Menu now shows that 2 wavelets are being used:

February, 2011 Click on Next and Ok on this menu. 300


Now, a new menu appears, allowing you to set the background relationship
between ln(ZP), ln(ZS), and ln(Density):

These cross plots have


been calculated using
the full range of the logs
from the AVO well. An
improved estimate could
be made by limiting the
depth range of the data
being used.

For now, we will simply


manually improve the
regression fit through
the clusters.

February, 2011 301


Using the mouse,
modify the
regression lines
from this:

To this:

And click Ok and


Save regression
coefficients:

February, 2011 302


The Pre-stack Inversion
Menu now looks like this
(note that the coefficients
may not be exactly the
same):

All the parameters on this


menu can be defaulted.

Click on Apply to see the


inversion result at the well:
February, 2011 303
The display shows a number of useful curves:

Real Log

Initial
Model

Inversion
Synthetic
Trace
Error
Real
February, 2011 304
Data
To customize the display, click on the “eyeball” icon:

On the Layout page, remove


the plot of Density and add
the plot of Vp/Vs:

Then, click on the Curves page:

And select the option to


Apply a filter to the
original logs for display.
Finally, click on Ok.

February, 2011 305


The new display shows a very good fit between the inversion traces and the original
logs, especially near the target zone:

February, 2011 306


Now we will apply the inversion to the entire data set.
Click on Inversion / Pre-stack Inversion:

The menu that follows confirms all


the parameters we have already
seen. So we can default every
page, except the last one, which
determines which volumes will be
created. Since we have chosen
NOT to update Density, we remove
it from the list and add Zs instead
Now, click Ok to create the
inversion volumes:

February, 2011 307


A series of windows now appears.
One window is the synthetic data
corresponding to the inversion
output.

We can also see the “error”, which


is the difference between the real
data and synthetic data.

Click on the “eyeball” and set the


Trace Data Volume as shown below:

February, 2011 308


The other volumes
are Zp, Zs, and
Vp/Vs. These
volumes are now
available for
visualization, cross
plotting or Emerge
analysis.

We have completed
the pre-stack
inversion project.

Close down the


STRATA program by
clicking on File / Exit
Project on any of the
STRATA windows.

February, 2011 309

Вам также может понравиться