This document summarizes a method for using potential field data and stochastic optimization to refine 3D geological models. [1] It demonstrates tools for gravity and magnetic modeling that incorporate forward modeling, property optimization, and geometry optimization. [2] A synthetic example and field example are provided to test the method. [3] The goal is to recover buried geological structures from potential field data by iteratively optimizing the geometry and properties of 3D geological models.
2. Outline of presentation
• Demonstrate a set of tools for gravity and
magnetic modelling
• Used to refine 3D geological maps
• Incorporating 3 complementary procedures
– Forward modelling
– Property optimisation
– Geometry optimisation
• Synthetic example (dipping slab)
• Field example (San Nicolas VMS deposit)
3. Workflow outline
Geological observations Mass density and magnetic Gravity and magnetic
property observations observations
3D geological map (geometry)
and bulk property estimates
Supply geometry
Supply geometry Supply properties
and properties
Forward model Optimise Optimise
properties geometry
Simulated Simulated Simulated
response response response
+ +
property revised geometry
estimates
Reject or revise the geological map and property estimates
4. “Slab” synthetic example
• Part 1
– Create a 3D geological map with a dipping slab
unit in a host unit
• e.g., VMS deposit in a host sequence
– Assign anomalous density and magnetic
properties to the slab
– Generate synthetic data
– Add random noise to form an “observed” data set
• Part 2
– Try to recover the source feature
5. Triangulated form Voxel form
Density Susceptibility Remanent Magnetisation
t/m3 SI x105 A/m
Slab 3.27 1000 0.5 ( I = 35°, D = 135° )
Host 2.67 10 0
Vertical gravity (gD) TMI ( I = -65°, D = 25° )
6. Surface geology (i.e., Slab not visible!)
Prior property estimates
Density Susceptibility Remanent Magnetisation
t/m3 SI x105 A/m
Slab 3.27 1000 ???
Host 2.67 10 0
Vertical gravity (gD) + noise TMI ( I = -65°, D = 25° ) + noise
7. Build an initial 3D geology map
• We observe discrete gD
and TMI anomalies
• Propose a VMS deposit
as the source
– Elevated density and
magnetisation Initial geological map in
triangulated surface form
relative to the host
• Build a simple 3D
geological map with a
buried deposit (vertical
slab geometry) in a host
unit
Discrete voxel form
8. Density Susceptibility Remanent Magnetisation
Prior property t/m3 SI x105 A/m
estimates Slab 3.27 1000 ???
Host 2.67 10 0
Vertical gravity (gD) TMI ( I = -65°, D = 25° )
Observed data
Forward model
data
9. a1 *
Response
for Property
geological
unit #1 optimisation
Response
for
+ a2 * geological
unit #2
…
Response
+ an * for
geological
unit #n
+ an+1 * Trend
= Observed
response
Optimise a1 to an+1 (which are the property contrasts)
10. Density Total effective magnetisation
Recovered t/m3 A/m
property Slab 3.76 (3.27) 0.63, I=-15°, D=115° (0.39, I=-11°, D=111°)
estimates Host 2.67 * ( Susceptibility 18 SI x105 )
(true values in Vertical gravity (gD) TMI ( I = -65°, D = 25° )
braces)
Observed data
Property
optimisation
data
(Assuming remanent
magnetisation of unknown
direction for Slab)
11. Start
Geometry
optimisation Load geological and property models
Propose changes to the models
Fail
Apply geological tests
Pass
Fail
Forward model and apply
geophysical tests
Pass
(Modifications rejected,
(Modifications accepted and saved)
revert to previous models)
Continue?
Finish
12. Boundary modification scheme
• Select at random a voxel that is on
a geological boundary
• Propose a change to the geological
assignment by randomly choosing
from the list of map units in the
neighbourhood of this voxel
• Re-sample properties according to
the distribution for the new map unit Present model
• Then …
– Apply geological tests
– Calculate geophysical response
– Apply geophysical tests
• And repeat the process (over and
over)
Proposed model
14. Geological reference model Vertical Section
750 m
(‘most probable’ prior model) 500 mN
2 km
Slab
Host
‘True’ model
( Vertical exaggeration 1:1 )
15. Animation of the geometry optimisation
process for Section 500 mN
( Vertical exaggeration 1:1, frames captured at increments of 500 proposals from 1 to 100,000 proposals )
16. Geological reference model Vertical Section
750 m
(‘most probable’ prior model) 500 mN
‘Most probable’
(posterior composite model)
2 km
‘Most probable’ thresholded Slab
(posterior composite model Host
with P ≥ 95 % )
Probability shown with logarithmic scaling
‘Probability’ for Slab 100 %
10 %
1 % (or less)
‘True’ model
( Vertical exaggeration 1:1 )
17. Comparison of geometry optimisation using
single and multiple data types
Vertical gravity (gD) TMI Joint gD and TMI
Probability shown with logarithmic scaling
Prior reference model
100 %
10 % True model
1 % (or less)
( Vertical section 500 mN, Vertical exaggeration 1:1 )
18. San Nicolas VMS deposit, Mexico
Vertical section -400 mN
Mafic Volcanics Tertiary Breccia
2000
175m
Elevation (m)
Quartz Massive Sulphide
Rhyolite
Mafic Volcanics
Basin and
Range el”
“ Ke
1600
Province
-2000 Easting (m) -1100
San Nicolas (Section from Phillips et al., 2001)
Cenozoic cover
Mesozoic
basement
(Section adapted from a figure supplied by the USGS)
19. Interpreting gravity and magnetic data
as part of the exploration of the region
• Purpose – To identify and perform preliminary
evaluation of VMS targets
• Data assessment
– Recognition of the importance of the Cenozoic
cover for generating the geophysical response
• Determine the thickness and properties of the
Cenozoic cover
• Select targets
• Assess the viability of the targets
• Provide information to assist drill testing of the viable
targets
20. Surface topography Surface geology Rock properties
Density Susc. Cond.
t/m3 105 SI mS/m
Cenozoic 2.17 10 – 400 40
Mesozoic
3.27 1000 50
(Sulphides)
Mesozoic (Mafic?) 2.72 100 N/A
Mesozoic 2.67 10 Low
AEM conductance Vertical gravity RTP TMI
12 km
Hot colours = high conductance
(Adapted from material supplied by various sources)
21. Cover / basement vertical gravity forward
modelling and property optimisation
Thickness of
the Cenozoic Observed gD
cover derived
from AEM
conductance
estimates
-4.0 to +0.5 mGal
0 to 250 m
Forward model Property
gD optimisation
gD
Cover 2.30 t/m3 Cover 2.18 t/m3
Basement 2.67 t/m3 * Basement 2.67 t/m3 *
-4.0 to +0.5 mGal -4.0 to +0.5 mGal
23. Section - 400 mN
2 km
Geological reference model (‘most probable’ prior model)
12 km
Cenozoic cover
Mesozoic basement
( Vertical exaggeration 1:1 )
24. Animation of the geometry optimisation
process for Section -400 mN
( Vertical exaggeration 1:1, frames captured at increments of 20,000 proposals from 1 to 8 million proposals )
25. Section - 400 mN
2 km
Geological reference model (‘most probable’ prior model)
12 km
‘Most probable’ posterior composite model
Cenozoic cover
Mesozoic basement
‘Most probable’ posterior composite model ( P ≥ 95 % )
Logarithmic scaling
100 %
10 %
‘Probability’ for Cenozoic cover 1 % (or less)
( Vertical exaggeration 1:1 )
26. Section - 400 mN
2 km
Cenozoic cover
Mesozoic basement
‘Most probable’ posterior composite model
Zero thickness
Thickness of Cenozoic cover
( i.e., basement outcrop or
shallow sub-crop )
( from geometry optimisation
of basement/cover model )
Contour interval 100 m
28. Initial selection of targets
Vertical gravity residual RTP TMI
Region for detailed joint gD and TMI geometry optimisation
( 1200 m E/W and 800 m N/S, 25 m cell size )
29. Geological reference model Section - 400 mN
1000 m
(‘most probable’ prior model)
1200 m
‘Most probable’
(posterior composite model)
12 km
‘Most probable’ thresholded Cenozoic cover
(posterior composite model Mesozoic basement (sulphides)
with P ≥ 95 % ) Mesozoic basement (mafic)
Mesozoic basement
Probability shown with logarithmic scaling
‘Probability’ for Mesozoic
100 %
basement (sulphides)
10 %
1% (or less)
Supplied drill section
(simplified)
( Vertical exaggeration 1:1 )
30. Features of this approach
• Geological units used as the primary variables
• Operates in 3D
• Integrates geological information from a 3D geological map with
gravity and magnetic modelling
• Any combination of gravity and magnetic data types
– Scalar, vector or tensor components
• Joint gravity and magnetic investigations
• Properties sampled from statistical distributions
• Induced susceptibility and remanent magnetisation
• Conditional uncertainty estimates as well as parameter
estimates
– Stochastic approach (i.e., statistical or probabilistic method)
generates many models that fit the data adequately
31. Acknowledgements
• Geoscience Australia
• Intrepid Geophysics
• BRGM
• Teck Cominco
• Nigel Phillips (UBC-GIF, Mira Geoscience)
• Groups that have supported the development of
GeoModeller
• For further information, contact Richard Lane at
Geoscience Australia (richard.lane@ga.gov.au)
PDAC 2008