Details of the Inversion
workflow
Des Fitzgerald
August 2007
Geology Constrained Inversion
Simple Slab Tutorial
Part 1 Background
A statistical approach
2007
Objectives
Demonstrate new approaches to
inversion
Use all available geology, gravity and
magnetics to constrain the result
Limit the infinite possibilities to a set that is
consistent with all the facts
Bring latest inversion & probability
thinking to geoscience
Constraining Your Geology
Geomodeller has a set of geological rules to
resolve and constrain the initial model
Contacts
Dip/Strike
Pile, erode/onlap
Fault network
Inversion does not make direct use of these.
Instead, it uses a framework of geological
units,
important for producing realistic property models
allows simultaneous inversion of multiple
geophysical data types
Commonality
Possessing like and interchangeable
characteristics
The number of voxels in common
between the proposed models and the
reference model is controlled
statistically using a Commonality
Same j
Commonalit y j
constraint
Ref j
Geological Commonality Misfit
Say you want 90%
Shape Ratio
"Shape" is the square-root of surface area
divided by the cube-root of volume (to
mimimize any scale dependence).
"ShapeRatio" is the Shape of the formation in
the proposal divided by the Shape of the
formation in the ReferenceModel.
Shape of a sphere = 2.2
No scale dependency
Shape of a cylinder
h = r, Shape = 2.42
h = 2r, Shape = 2.35
h = 4r, Shape ~ 2.41
h = 20r, Shape ~ 2.89
Some aspect ratio dependency !
Cube
Shape ~ 2.45
Prism
square base r*r , height r/4 - Shape ~ 2.75
Geophysical Grid Preparation
Detrending
Need to be able to fit the anomalies
caused by signal in your model
External influences or deeper sources are
not relevant
Geomodeller does this on demand,
either just once at the beginning
Or on a regular basis as inversion
continues
Grids continued
You create a Measured Grids List
Type, (eg Gravity)
Mean Elev., (eg 50m )
Precision, (eg 0.5 mGals)
Detrend, (eg Yes/No)
Detrend degree, (eg 0,1,2 etc)
detrend freq., (eg 0, or 100000)
grid URL (eg c:/data/[Link])
Geophysical Precision
An estimation of the standard deviation
of the measurement error in your
geophysical data
All observed data has inherent error due
to a variety of factors
Typically, ground gravity would be good
to about 0.5 mGals
Airborne Total Magnetic Intensity to
about 1 nT
Overview of Cases
1. Forward Model
Propose a start model, compute gravity response
2. Prior Only
Propose a geological model, explore probability
space implied by the geological constraints
3. Fixed Geometry
Solve for properties using a bounded least-squares
fit
4. Bi-Modal Property Only
No proposed body, just a proposed density contrast
5. Constrained Geology
Use gravity to constrain result
Overview continued
7. Constrained Geology
Use magnetics assuming induced response only
8. Constrained Geology
Joint gravity and magnetics
9. Constrained Geology
Use magnetics with the possibility of remanence
10. Constrained Geology
Use observed tensor gradients
Property Uncertainty
vs
Geological Modelling
By default, we give equal importance to
these two differing aspects
You can speed convergence of your
inversion, if you have a high confidence
in your properties and there in no
spread of values
Simply adjust all the effort towards the
geological uncertainty
Tools & Status
The Geomodeller Tool as used for creating
your 3D geological model, can also perform
Forward Modelling functions
The ManageLithoInversion Tool is a Batch
Processing tool only at V1.2
This is used for all other aspects of Inversion
Voxets
Can be examined by Voxler, other thirdparty tools
Movies
Standard thirdparty products
Geology Constrained Inversion
Simple Slab Tutorial
Part 2 Practical
A statistical approach
2007
Test Problem
Find Buried Object
Unknown dip / strike
Unknown density contrast
Unknown magnetic
susceptibility
Observed Gravity
Observed Geophysics
Vertical component of gravity
Total magnetic intensity
No remanence
Observed Magnetics
Make initial Geology Model
Propose a simple
vertical slab as a
starting model
Buried 100 m
Dimension of 100 *
200 * 400
Make model
Save project
Initial Plan view of model
200m below surface
Test Properties
Slab density 3.67
Host density 2.67
Slab susceptibility 0.001
Host susceptibility 0.000001
Optional Extras
Slab remanence (0.05,0.00,100,25,115)
Host remanence (0.00,0.00,100,0,00)
Geophysics Observations
Data Collection height : 50m
Gravity Data
normal ground observations
Magnetic Data
Field Intensity: 50000
Inclination: -65
Declination: 25
Conditioning the Geophysics
For both Gravaity and Magnetics, we
recommend you turn detrending ON
A first order trend should be fine, so
choose 1
Gravity detrending should not need to
be repeated as the inversion evolves
Sometimes Magnetic data can benefit
from a further detrending adjustment of
the misfit say every 100000 iteration
Actual Body
Preparing a case
NewCase Command
Create an Inversion Case
Within this case the following are
treated as constant
The voxet resolution and extent
The reference geological model
The physical property laws
The voxels to hold fixed
The geological constraints
The observed geophysical grids
Preparing a Run
Create an Inversion Run
Within this run the following are treated
as constant
The Case
The number of iterations
The initial lithology and physical properties
The starting or initial geological model
can be varied
Case 1 - Forward Modelling
Before starting the inversion, run
forward model first
This is done via the Geomodeller Tool
This way we check that the starting
geological model is reasonable
The assumed physical properties can
be checked
Case 1 Instructions
Start Geomodeller
Load Simple slab
Geophysics menu
3d Forward/Inversion
Choose G00 (gravity)
Constant grid
Choose apply
Gravity Grids
Observed
Initial
Forward
Not bad but missing some plunge??
Case 2 Prior Only
This is always recommended before full inversion
Do you have good geological control of proposed
models ie are your geological models viable?
Methodology
Ignore the geophysics
Constrain the geological arrangements with respect to a
reference geology
Randomly perturb, evaluate model geologically using one or
more of:
1. Shape ratio constraint
2. Commonality of new with reference
3. Volume ratio
Prior Only is an expression of geological uncertainty
PRIOR ONLY
computational scheme
Start
No Geophysics used
Load start litho- & property models
Modify models
Fails
Fails
(Modification rejected)
Test against geology model criteria
Test against geophysics dataset
criteria
(Modification accepted)
Continue?
Finish
Mechanics
You describe your process via the file :
[Link]
This is both human and machine readable
Try internet explorer on it!
A copy of Your Geomodeller Project is also made for
each case this is also mostly an XML file
Current interface is via batch scripts
An inversion run can take time to complete
A voxet of 50m x 50m x 50m is produced
A voxel change log for every viable model is kept so
that it is possible to reconstruct the evolving state
after the event.
[Link]
<InversionState Version="3" xmlns:xsi="[Link]
- <Case AboveTopoDensityLaw="Normal(2.67,0,100)"
AboveTopoRemanenceLaw="Normal(0,0,100,0,0)" AboveTopoSusceptibilityLaw="Normal(0,0,100)"
AmbientMagneticDeclination="25
AmbientMagneticInclination="-65"
AmbientMagneticMagnitude="50000"
DensityRef="2.67" Dx="50" Dy="50" Dz="50" InPlace="false" IncludeBorderEffect="true"
Name="Case_7" ReferenceModelPath="Case_7.vo" UseTopo="true" VoxetPath="Case_7.vox"
WorkDir=".." Xmax="975" Xmin="25" Ymax="975" Ymin="25" ZmaxRequested="0" ZmaxVoxet="-25"
ZminRequested="-500" ZminVoxet="-475">
<Lithology CommonalityLaw="Weibull(0.05,1.000000)"
CommonalityVolumeLaw="LogNormal(0.000000,1.000000)"
DensityLaw="Normal(2.67,0.00001,100)" Formation="Host" Index="1" Movable="true"
RemanenceLaw="Normal(0,0,100,0,0)"
ShapeRatioLaw="LogNormal(0.050000,0.050000)"
SusceptibilityLaw="LogNormal(0.0000001,0.0000001,100)"
VolumeRatioLaw="LogNormal(0.000000,0.010000)" />
<Lithology CommonalityLaw="Weibull(0.02,1.000000)"
CommonalityVolumeLaw="LogNormal(0.000000,1.000000)"
DensityLaw="Normal(3.67,0.00001,100)" Formation="Slab" Index="2" Movable="true"
RemanenceLaw="Normal(0,0,100,0,0)"
ShapeRatioLaw="LogNormal(0.100000,0.050000)"
SusceptibilityLaw="LogNormal(0.0010000,0.0000001,100)"
VolumeRatioLaw="LogNormal(0.000000,0.010000)" />
Run Parameters 1
Iterations 100000
Commonality: Y
Slab Commonality Weibull(0.02,1.0)
Host Commonality Weibull(0.05,1.0)
Shape Ratio: Y
Slab ShapeRatio LogNormal(0.1,0.05)
Host ShapeRatio LogNormal(0.0,0.05)
Geology Constrained Inversion
Simple Slab Tutorial
Part 3 Reporting
A statistical approach
2007
Recommended Discipline for
Project layout
Querying Inversion Outputs
Once the inversion itself is complete various
visuals can be produced to analyse the
results
Images along sections
Movies along sections
Geophysical grids of the computed geophysics
Surface mesh representing isovalues
Probability of lithology voxets
Available Queries
MakeSummaryStats
Work out the summary statistics of all models between a start and
end iteration. From this we can derive products such as the
probability of a given lithology at a given cell
MakeHistogram
Generates a histogram of densities and/or susceptibilities for each
Formation in a Voxel dataset
MakeSectionImage
Generates a section through a Voxel dataset and presents it as a jpg
image that can then be displayed from within 3D GeoModeller
MakeDerivedVoxet
Various operations on Voxel datasets. Ex. Searches though a list of
Sujmary stats datasets and can find the most probable formation
MakeEvolutionMovie
On a nominated section can create an avi movie of the evolution of
the lithology model though the iterations
Example of 2 SuperSummary Stats Outputs
The names of the various queried output properties are defined in the voxet
header: eg 85% & 95% probabilty effects only property 5 below.
SuperSummaryStats_Case_4_Run_1_iterations_3000000_1000000_85_super.vo,
AND identically also in:
SuperSummaryStats_Case_4_Run_1_iterations_3000000_1000000_95_super.vo
Property 1 = Change Count
Property 2 = Mean Density
Property 3 = Std Dev Density
Property 4 = Most Probable
Property 5 = Most Probable Thresholded
Property 6 = probability 0 (First lithology from model)
Property 7 = probability 1 (Second lithology from model)
Property 8 = probability 2
Property 9 = probability 3
Property 10 = probability 4
Property 11 = probability 5
Property 12 = probability 6
Each type of output statistic generated by MakeSuperSummaryStats
Definition: Most probable thresholded
means
Example section
For the entire inversion after burn in, and considering
all of the accepted, proposed geology models
Report back all the voxets which remain a single
lithology for a given %, or more, of the total # of
proposals.
Report back the location and lithology of those
voxets.
Black volumes indicate areas of greater changeability
of lithology, such that the threshold % was not met.
Section 130_000_N: Most probable geology
Section 130_000_N: Most probable geology Thresholded 85%
Section 130_000_N: Most probable geology Thresholded 95%
ProbabilityResults
Run1 - All random
changes allowed
Run 2 - Body
constrained by
commonality and
shape ratios
Mean Density
Predicted
density
distributions
based purely
on geology
constraints
Raw Statistics of Prior Only Run
AcceptanceCount="73365
AcceptanceRatePerHour="434214"
CommonalityRejectCount="12969"
CommonalityTestCount="99999"
ComputerName="RAYS-PC"
Counter="100000"
FinishedDateTime="24/ 8/2007 9:24:18.122000"
InterAcceptanceDifferentFormationCount="8926"
InterAcceptanceSameFormationCount="64439"
InterProposalDifferentFormationCount="35560"
InterProposalSameFormationCount="64439
ProposalRatePerHour="591849"
ShapeRejectCount="13665"
ShapeTestCount="87030"
StartedDateTime="24/ 8/2007 9:14:9.443000
Super Summary Stats
Statistical History of each voxel
Change count
Mean property
Standard deviation
Probability for each lithology
Most probable lithology over certain
threshhold ( say > 90%)
Visualise via a voxet viewer or sectional
images or movies
Run Parameters 2
We can show how the predicted bodies can be
tightened up to honour the original shapes by
changing the commonality
Iterations 100000
Commonality: Y
Slab Commonality Weibull(0.1,1.0)
Host Commonality Weibull(0.1,1.0)
Shape Ratio: Y
Slab ShapeRatio LogNormal(0.10,0.05)
Host ShapeRatio LogNormal(0.0,0.05)
Case 3 - No Geology Body Shape
Bi-Modal Properties
Another case to show the tool working
in a way that is closest to existing
deterministic methods
Prepare a voxet with no buried geological
body
Propose the existing of a second
population via a bi-modal property law
Create Physical Properties 1
Create Bi-Modal Density for Host
80%
background
20% more
dense
Gravity Inversion of Bi-Modal
Case
Final Density Section Image
Since we
have no
geological
control the
density
anomaly is
biased to the
near surface
NB takes
500000
iterations
Case 4 - Gravity Inversion
Enforce commonality
Enforce Shape Ratio
Use observed ground gravity
Run for 100000 iterations
Gravity plus geology
constraints
computational scheme
Start
Load start litho- & property models
Modify models
Fails
Fails
(Modification rejected)
Test against geological model
Test against gravity dataset
(Modification accepted)
Continue?
Finish
Dipping Slab
Gravity Grids
Observed
Initial
Final
Case 5 Magnetics Inversion
Assume a small vertical slab body as a
starting model
Constrain the geology via commonality
and shape ratio
Use observed magnetic data to refine
aspects of the bodys geometry and
properties
Assume the magnetics is induced only
Dipping Slab
Magnetic Grids
Observed
Initial
Final
Start
Overview of the
computational
scheme
Load start litho- & property models
Modify models
Fails
Fails
(Modification rejected)
Test against model criteria
Test against dataset criteria
(Modification accepted)
Continue?
Finish
Case 6 - Joint Inversion
We have shown the dipping slab case
using the observed gravity
Now we add observed magnetics to our
case and run joint inversion
Run for 300000, with a burn-in of
200000
Most Probable Lithology
Joint Miss-fit evolution
Gravity and magnetic misfit during inversion
10
misfit (nT/mGal)
6
gravity
magnetic
0
0
1000000
2000000
3000000
# iterations
4000000
5000000
6000000
Comments
This shows good adaption to the
expected outcome
Both Mag. And Gravity contribute
Over-view of Outcomes
Gravity Only inversion
does not do depth estimates well
Volume/mass of geological anomaly is quite good
Induced Magnetics inversion
Depth to top of body is good
Orientation of body is also good
Poor estimate of body volume
Joint Magnetic & Gravity
Get both depth and volume OK
