3d potential field inversion for wireframe surface geometryfarq/pdfs/peter_seg_2015.pdf · 3d...
TRANSCRIPT
3D Potential Field Inversion forWireframe Surface Geometry
Peter G. Lelievre1, Colin G. Farquharson1 and Rodrigo Bijani2
http://www.esd.mun.ca/~peter/Home.html
http://www.esd.mun.ca/~farq
1Memorial University, Department of Earth Sciences, St. John’s, Canada2Observatorio Nacional, Rio de Janeiro, Brasil
SEG 85th Annual Meeting, Oct. 2015, New Orleans, GM2
Introduction Mesh-based inversion Surface-based inversion (2D) Surface-based inversion (3D) Conclusion
Geophysical inversion primer
Earth model (e.g. density) Survey data (e.g. gravity)
Forward problem
Inverse problem
Lelievre1, Farquharson1 and Bijani2, [email protected] 1Memorial University, 2Observatorio Nacional
3D potential field inversion for wireframe surface geometry (2 / 27)
Introduction Mesh-based inversion Surface-based inversion (2D) Surface-based inversion (3D) Conclusion
Motivation
(a) (b)
• Geophysical numerical methods typically work with mesh-baseddistributions of physical properties (a)
• Geologists’ interpretations about the Earth typically involve wireframecontacts between distinct rock units (b)
• There is a disconnect here!
Lelievre1, Farquharson1 and Bijani2, [email protected] 1Memorial University, 2Observatorio Nacional
3D potential field inversion for wireframe surface geometry (3 / 27)
Introduction Mesh-based inversion Surface-based inversion (2D) Surface-based inversion (3D) Conclusion
Types of geophysical inversion
1 Discrete body inversion
2 Mesh-based inversion
3 Surface-based inversion
Lelievre1, Farquharson1 and Bijani2, [email protected] 1Memorial University, 2Observatorio Nacional
3D potential field inversion for wireframe surface geometry (4 / 27)
Introduction Mesh-based inversion Surface-based inversion (2D) Surface-based inversion (3D) Conclusion
1. Discrete body inversion
Simplified representation of the Earth:• Simple shapes for one or more causative
target bodies
• Homogeneous background
Inversion:
• Few parameters (e.g. shape, location)
• Data best-fit problem
• Low computational requirements
• Stochastic investigations feasible
Lelievre1, Farquharson1 and Bijani2, [email protected] 1Memorial University, 2Observatorio Nacional
3D potential field inversion for wireframe surface geometry (5 / 27)
Introduction Mesh-based inversion Surface-based inversion (2D) Surface-based inversion (3D) Conclusion
2. Mesh-based inversion
General representation of the Earth:• Mesh of tightly packed cells
• Piecewise (pixellated) distributionof physical properties
Inversion:
• Many parameters (many cells)
• High computational requirements
• Stochastic investigations not veryfeasible
Rectilinear Quadtree
Unstructured
Lelievre1, Farquharson1 and Bijani2, [email protected] 1Memorial University, 2Observatorio Nacional
3D potential field inversion for wireframe surface geometry (6 / 27)
Introduction Mesh-based inversion Surface-based inversion (2D) Surface-based inversion (3D) Conclusion
2. Mesh-based inversion
General representation of the Earth:• Mesh of tightly packed cells
• Piecewise (pixellated) distributionof physical properties
Inversion:
• Many parameters (many cells)
• High computational requirements
• Stochastic investigations not veryfeasible
Lelievre1, Farquharson1 and Bijani2, [email protected] 1Memorial University, 2Observatorio Nacional
3D potential field inversion for wireframe surface geometry (6 / 27)
Introduction Mesh-based inversion Surface-based inversion (2D) Surface-based inversion (3D) Conclusion
3. Surface-based inversion
Flexible representation of the Earth:• Wireframe of nodes and
connecting facets representingcontacts between rock units
• How geological models are built
Inversion:
• Surface geometry defined bymoderate number of parameters
• Moderate computationalrequirements
• Stochastic investigationssomewhat feasible
Richardson & MacInnes, 1989, The inversion of gravity datainto three-dimensional polyhedral models, JGR
Lelievre1, Farquharson1 and Bijani2, [email protected] 1Memorial University, 2Observatorio Nacional
3D potential field inversion for wireframe surface geometry (7 / 27)
Introduction Mesh-based inversion Surface-based inversion (2D) Surface-based inversion (3D) Conclusion
3. Surface-based inversion
Flexible representation of the Earth:• Wireframe of nodes and
connecting facets representingcontacts between rock units
• How geological models are built
Inversion:
• Surface geometry defined bymoderate number of parameters
• Moderate computationalrequirements
• Stochastic investigationssomewhat feasible
Lelievre1, Farquharson1 and Bijani2, [email protected] 1Memorial University, 2Observatorio Nacional
3D potential field inversion for wireframe surface geometry (7 / 27)
Introduction Mesh-based inversion Surface-based inversion (2D) Surface-based inversion (3D) Conclusion
3. Surface-based inversion
Flexible representation of the Earth:• Wireframe of nodes and
connecting facets representingcontacts between rock units
• How geological models are built
Inversion:
• Surface geometry defined bymoderate number of parameters
• Moderate computationalrequirements
• Stochastic investigationssomewhat feasible
Lelievre1, Farquharson1 and Bijani2, [email protected] 1Memorial University, 2Observatorio Nacional
3D potential field inversion for wireframe surface geometry (7 / 27)
Introduction Mesh-based inversion Surface-based inversion (2D) Surface-based inversion (3D) Conclusion
Types of geophysical inversion
1 Discrete body inversion
2 Mesh-based inversion
3 Surface-based inversion
Lelievre1, Farquharson1 and Bijani2, [email protected] 1Memorial University, 2Observatorio Nacional
3D potential field inversion for wireframe surface geometry (8 / 27)
Introduction Mesh-based inversion Surface-based inversion (2D) Surface-based inversion (3D) Conclusion
Mesh-based inversion for smooth distributions
• Objective function
Φ = Φd + βΦm
• Data misfit
Φd =∑i
(F (m)i − di
σi
)2
• Model structure (regularization)
Φm =∑j
wj(mj − pj)2 +
∑j
∑k
wj ,k(mj −mk)2
[smallness term] + [smoothness term]
• Deterministic local optimization approach: one “best” solution
Lelievre1, Farquharson1 and Bijani2, [email protected] 1Memorial University, 2Observatorio Nacional
3D potential field inversion for wireframe surface geometry (9 / 27)
Introduction Mesh-based inversion Surface-based inversion (2D) Surface-based inversion (3D) Conclusion
Mesh-based inversion for sharper features
• Objective function
Φ = Φd + βΦm
• Data misfit
Φd =∑i
(F (m)i − di
σi
)2
• Model structure (regularization)
Φm =∑j
wj(mj − pj)2 +
∑j
∑k
wj ,k |mj −mk |p + Ψ
[smallness term] + [smoothness term]
• Different norm, measures or re-weighted iterative procedure can help
Lelievre1, Farquharson1 and Bijani2, [email protected] 1Memorial University, 2Observatorio Nacional
3D potential field inversion for wireframe surface geometry (10 / 27)
Introduction Mesh-based inversion Surface-based inversion (2D) Surface-based inversion (3D) Conclusion
Mesh-based inversion for sharper features
• Objective function
Φ = Φd + βΦm
• Data misfit
Φd =∑i
(F (m)i − di
σi
)2
• Model structure (regularization)
Φm =∑j
wj(mj − pj)2 +
∑j
∑k
wj ,k(mj −mk)2
[smallness term] + [smoothness term]
• The safest and most effective approach is to hardwire the surfaces
Lelievre1, Farquharson1 and Bijani2, [email protected] 1Memorial University, 2Observatorio Nacional
3D potential field inversion for wireframe surface geometry (10 / 27)
Introduction Mesh-based inversion Surface-based inversion (2D) Surface-based inversion (3D) Conclusion
Mesh-based inversion for sharper features
Hardwiring surfaces ⇒ unstructured meshes become important
Inversion of gravity gradiometry data:
Unconstrained Hardwired surface Additional regularization
Lelievre1, Farquharson1 and Bijani2, [email protected] 1Memorial University, 2Observatorio Nacional
3D potential field inversion for wireframe surface geometry (11 / 27)
Introduction Mesh-based inversion Surface-based inversion (2D) Surface-based inversion (3D) Conclusion
Types of geophysical inversion
1 Discrete body inversion
2 Mesh-based inversion
3 Surface-based inversion (2D)
Lelievre1, Farquharson1 and Bijani2, [email protected] 1Memorial University, 2Observatorio Nacional
3D potential field inversion for wireframe surface geometry (12 / 27)
Introduction Mesh-based inversion Surface-based inversion (2D) Surface-based inversion (3D) Conclusion
2D Surface-based inversion for geometry
• Inversion seeks positions of nodes inwireframe model
• Only data misfit is required
Φd =∑i
(F (m)i − di
σi
)2
• Regularization not required: work on coarserepresentation, refine e.g. splines
• Global optimization strategies (PSO, GA, MCMC) provide statistics:
⇒ many solution samples
Lelievre1, Farquharson1 and Bijani2, [email protected] 1Memorial University, 2Observatorio Nacional
3D potential field inversion for wireframe surface geometry (13 / 27)
Introduction Mesh-based inversion Surface-based inversion (2D) Surface-based inversion (3D) Conclusion
Cocagne Subbasin: gravity survey data
CALEDONIAUPLIF
T
MONCTONSUBBASIN
HASTINGS UPLIFT
WESTMORLAND UPLIFT
SACKVILLESUBBASIN
NEWBRUNSW
ICK
PLATFORM
BelleisleFault
COCAG
NE
SUBBASIN
Smith
Creek
Fault Nor
th River
Fault
ZONE
DEFORMED
INDIA
N
MO
UNTAIN
Sussex
Chi
gnec
toB
ay
NovaScotia
Northumberland
Strait
Alma
Bayof
Fundy
Shediac
Sackville
CUM
BERLA
ND
SUBBASIN
0 10 km
0 10 km
66°
U.S.A.
NewBrunswick
Nova
Scotia
P.E.I.
Gulfof
St. Lawrence Newfoundlandand Labrador
Qu becé
0 100 km
Area of Figure 1
MARITIMESBASIN
Buctouche
Richibucto
BerryMillsFault
Moncton
Area ofFigure 2
Late Devonian–Carboniferousdeep basins
Pre-Late Devonian crystallinebasement
Carboniferous shallow basins
Stoney Creek Field (oil/gas)
McCully Field (gas)
Penobsquis mine (potash/salt)
Fault
Figure 1. Distribution of Late Devonian–Carboniferous uplifts and subbasins in theMaritimes Basin of eastern New Brunswick(modified after St. Peter 2006; St. Peter andJohnson 2009), as well as the location ofpresent-day oil, natural gas, and potash/saltproducers. Inset map shows the area of theMaritimes Basin in eastern Canada.
101
(previously interpreted trace)
113
65°10’ 65°00’ 64°50’ 64°40’ 64°30’
46°40’
46°30’
46°20’
46°10’
Figure 8. New terrain-corrected Bouguer anomaly map of the report area, with fBlue lines E–E' and W–W' mark the location of
section profiles in Figures 11 and 12, respectively.
ault boundaries(modified after Smith 2008) shown as thick grey lines.
Gravity stations are as identified in Figure 6. Figure 1shows the location of this report area in eastern New Brunswick.
0 5 10 km
W
W'
E'
E
BelleisleFault
CormiervilleFault
Smith CreekFault
Berry MillsFault
J. Evangelatos & K. E. Butler, 2010. New gravity data in Eastern New Brunswick: implications for structural delineation ofthe Cocagne Subbasin. In Geological Investigations in New Brunswick for 2009. Edited by G. L. Martin. New BrunswickDepartment of Natural Resources; Lands, Minerals and Petroleum Division, Mineral Resource Report 2010–1, p. 98–122.
Lelievre1, Farquharson1 and Bijani2, [email protected] 1Memorial University, 2Observatorio Nacional
3D potential field inversion for wireframe surface geometry (14 / 27)
Introduction Mesh-based inversion Surface-based inversion (2D) Surface-based inversion (3D) Conclusion
Cocagne Subbasin: rudimentary geological model
Figure 11. A two-dimensional geological model, simulating the observed gravity response across theeastern part of the Cocagne Subbasin along profile E–E' in Figure 8. The magnetic profile is extractedfrom the regional aeromagnetic dataset (Canadian Belleisle,Cormierville, and Smith Creek fault traces in Figure 8 are shown here as red ticks on the horizontal axis.
Aeromagnetic Data Base 2010). The
Belleisle Fault4
0 5 10 15
Smith Creek Fault
Basement rock(2700 kg/m )3
Basement rock(2710 kg/m )3
Sedimentaryrock (2550 kg/m )3
Cormierville Fault
3
2
1
Depth(km)
0
E (North) E' (South)
-5
0
5
10
600
400
200
0
Gravity(mGal)
Magnetics(nT)
Distance (km)
VerticalExaggeration = 2
Modelled
Observed
Observed
118
constraints on subsurface densities might be obtained by consulting well logs for the fewmoderately deep boreholes in the region, although none penetrates deeply into the Cocagnegravity low itself.
Structural constraints might also be obtained by analyzing old seismic reflection data from thearea. Indeed, Durling and Marillier (1990a) discuss a structural interpretation of marineseismic reflection data acquired from the nearby offshore area in the Northumberland Straitand Gulf of St. Lawrence. Further information about correlations between the onshore and
J. Evangelatos & K. E. Butler, 2010. New gravity data in Eastern New Brunswick: implications for structural delineation ofthe Cocagne Subbasin. In Geological Investigations in New Brunswick for 2009. Edited by G. L. Martin. New BrunswickDepartment of Natural Resources; Lands, Minerals and Petroleum Division, Mineral Resource Report 2010–1, p. 98–122.
Lelievre1, Farquharson1 and Bijani2, [email protected] 1Memorial University, 2Observatorio Nacional
3D potential field inversion for wireframe surface geometry (15 / 27)
Introduction Mesh-based inversion Surface-based inversion (2D) Surface-based inversion (3D) Conclusion
Cocagne Subbasin: mesh-based inversion
−15
−10
−5
0
5
10
Gra
vity (
mG
al)
0 5000 10000 15000
Distance (m)
ObservedModelled
−15
−10
−5
0
5
10
Gra
vity (
mG
al)
0 5000 10000 15000
Distance (m)
ObservedPredicted
Reference model Recovered model0
1
2
3
4
De
pth
(km
)
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Distance (km)
−0.2 −0.1 0.0 0.1
Anomalous density (kg/m3)
0
1
2
3
4
De
pth
(km
)
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Distance (km)
−0.2 −0.1 0.0 0.1
Anomalous density (kg/m3)
Lelievre1, Farquharson1 and Bijani2, [email protected] 1Memorial University, 2Observatorio Nacional
3D potential field inversion for wireframe surface geometry (16 / 27)
Introduction Mesh-based inversion Surface-based inversion (2D) Surface-based inversion (3D) Conclusion
Cocagne Subbasin: mesh-based inversion
−15
−10
−5
0
5
10
Gra
vity (
mG
al)
0 5000 10000 15000
Distance (m)
ObservedModelledPredicted
Absolute difference Percent difference0
1
2
3
4
De
pth
(km
)
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Distance (km)
0.00 0.01 0.02 0.03 0.04 0.05
Absolute density difference (kg/m3)
0
1
2
3
4
De
pth
(km
)
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Distance (km)
0 10 20 30 40 50
Percent density difference
Lelievre1, Farquharson1 and Bijani2, [email protected] 1Memorial University, 2Observatorio Nacional
3D potential field inversion for wireframe surface geometry (17 / 27)
Introduction Mesh-based inversion Surface-based inversion (2D) Surface-based inversion (3D) Conclusion
Cocagne Subbasin: surface-based inversion
−15
−10
−5
0
5
10
Gra
vity (
mG
al)
0 5000 10000 15000
Distance (m)
ObservedModelledPredicted
−15
−10
−5
0
5
10
Gra
vity (
mG
al)
0 5000 10000 15000
Distance (m)
ObservedModelledPredicted
2D wireframe model Splined model0
1
2
3
4
De
pth
(km
)
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Distance (km)
0.04 g/cc
0.03 g/cc
−0.12 g/cc
0
1
2
3
4
De
pth
(km
)
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Distance (km)
0.04 g/cc
0.03 g/cc
−0.12 g/cc
Lelievre1, Farquharson1 and Bijani2, [email protected] 1Memorial University, 2Observatorio Nacional
3D potential field inversion for wireframe surface geometry (18 / 27)
Introduction Mesh-based inversion Surface-based inversion (2D) Surface-based inversion (3D) Conclusion
Options for visualizing uncertaintyEnsemble of solutions Error bars (1 st. dev.)
0
1
2
3
4
De
pth
(km
)
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Distance (km)
0.04 g/cc
0.03 g/cc
−0.12 g/cc
0
1
2
3
4
De
pth
(km
)
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Distance (km)
0.04 g/cc
0.03 g/cc
−0.12 g/cc
Sized by st. dev. Coloured by st. dev.0
1
2
3
4
De
pth
(km
)
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Distance (km)
0.04 g/cc
0.03 g/cc
−0.12 g/cc
0
1
2
3
4
De
pth
(km
)
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Distance (km)
0.04 g/cc
0.03 g/cc
−0.12 g/cc
0 500 1000
Standard deviation
Lelievre1, Farquharson1 and Bijani2, [email protected] 1Memorial University, 2Observatorio Nacional
3D potential field inversion for wireframe surface geometry (19 / 27)
Introduction Mesh-based inversion Surface-based inversion (2D) Surface-based inversion (3D) Conclusion
Types of geophysical inversion
1 Discrete body inversion
2 Mesh-based inversion
3 Surface-based inversion (3D)
Lelievre1, Farquharson1 and Bijani2, [email protected] 1Memorial University, 2Observatorio Nacional
3D potential field inversion for wireframe surface geometry (20 / 27)
Introduction Mesh-based inversion Surface-based inversion (2D) Surface-based inversion (3D) Conclusion
Surface-based inversion for geometry
• Inversion seeks positions of nodes inwireframe model
• Only data misfit is required
Φd =∑i
(F (m)i − di
σi
)2
• Regularization not required: work on coarserepresentation, refine e.g. surface subdivision
• Global optimization strategies (PSO, GA, MCMC) provide statistics:
⇒ many solution samples
Lelievre1, Farquharson1 and Bijani2, [email protected] 1Memorial University, 2Observatorio Nacional
3D potential field inversion for wireframe surface geometry (21 / 27)
Introduction Mesh-based inversion Surface-based inversion (2D) Surface-based inversion (3D) Conclusion
Surface-based inversion for geometry
Use coarse control surfaces rather than the surfaces themselves!
Subdivision with cubicB-spline interpolation
Subdiv., Dyn-Levin-Gregory interp.
Subdiv., cubic B-spline
Bezier surface
Wojciech mula at pl.wikipedia. Licensed under CCBY-SA 3.0 via Wikimedia Commons
Lelievre1, Farquharson1 and Bijani2, [email protected] 1Memorial University, 2Observatorio Nacional
3D potential field inversion for wireframe surface geometry (22 / 27)
Introduction Mesh-based inversion Surface-based inversion (2D) Surface-based inversion (3D) Conclusion
Synthetic example #1: Isolated bodyBlack wireframe = true modelRed wireframe = control surface for true modelGreen wireframe = control surface for recovered modelWhite box = bounds on control nodes during inversionColoured surface = recovered model (standard deviations, red high)Coloured points = gravity data
Overhead view Side view
Lelievre1, Farquharson1 and Bijani2, [email protected] 1Memorial University, 2Observatorio Nacional
3D potential field inversion for wireframe surface geometry (23 / 27)
Introduction Mesh-based inversion Surface-based inversion (2D) Surface-based inversion (3D) Conclusion
Synthetic example #2: Sheet-like contact surfaceBlack wireframe = true modelRed wireframe = control surface for true modelGreen wireframe = control surface for recovered modelWhite box = bounds on control nodes during inversionLight grey surface = recovered model (standard deviations not plotted here)Coloured points = gravity data
Overhead view Side view
Lelievre1, Farquharson1 and Bijani2, [email protected] 1Memorial University, 2Observatorio Nacional
3D potential field inversion for wireframe surface geometry (24 / 27)
Introduction Mesh-based inversion Surface-based inversion (2D) Surface-based inversion (3D) Conclusion
Real data example: IOCG deposit, gravity data
Overhead view Side view
Surface-based inversion result consistent with understanding of geology,significant differences to mesh-based result require further study
Lelievre1, Farquharson1 and Bijani2, [email protected] 1Memorial University, 2Observatorio Nacional
3D potential field inversion for wireframe surface geometry (25 / 27)
Introduction Mesh-based inversion Surface-based inversion (2D) Surface-based inversion (3D) Conclusion
Summary
• Mesh-based “distribution inversion” is standard and numericallywell-behaved until you try to recover sharp features (best way is tohardwire them)
• Surface-based “geometry inversion” is challenging but:• can model sharp contacts and provide statistical information• geological and geophysical models can be, in essence, the same Earth
model: there is no longer a disconnect!
Lelievre1, Farquharson1 and Bijani2, [email protected] 1Memorial University, 2Observatorio Nacional
3D potential field inversion for wireframe surface geometry (26 / 27)
Introduction Mesh-based inversion Surface-based inversion (2D) Surface-based inversion (3D) Conclusion
Future work
• Joint surface-based inversion• Multi-objective optimization methods for joint inverse problems ...
• Alternate global optimization approaches• Particle swarm, genetic algorithms, ant colony, ...
• Work directly with complicated 3D common Earth models ...
• Hybrid approach ...
Lelievre1, Farquharson1 and Bijani2, [email protected] 1Memorial University, 2Observatorio Nacional
3D potential field inversion for wireframe surface geometry (27 / 27)
(additional slides follow)
Future work
Multi-objective genetic algorithms for joint inversion
Single-objective GA:• Aggregate of objectives:
min(f ) f = f1 + λf2 + ...
• One single best solution
• Difficult to find best λ value(s)
Multi-objective GA:
• Objectives treated separately:min(f1, f2)
• Several solutions along the Paretofront (nondominated)
Lelievre1, Farquharson1 and Bijani2, [email protected] 1Memorial University, 2Observatorio Nacional
3D potential field inversion for wireframe surface geometry (29 / 32)
Future work
Building and manipulating complicated 3D models
FacetModeller
Lelievre1, Farquharson1 and Bijani2, [email protected] 1Memorial University, 2Observatorio Nacional
3D potential field inversion for wireframe surface geometry (30 / 32)
Future work
Can we extract the control surfaces, rather than thesurfaces themselves, from the modelling software?
Subdiv., cubic B-spline Bezier surface
Wojciech mula at pl.wikipedia. Licensed under CCBY-SA 3.0 via Wikimedia Commons
Lelievre1, Farquharson1 and Bijani2, [email protected] 1Memorial University, 2Observatorio Nacional
3D potential field inversion for wireframe surface geometry (31 / 32)
Future work
A third approach?
1 Mesh-based inversion with sharp features
2 Surface-based inversion
3 Hybrid approach?Lelievre1, Farquharson1 and Bijani2, [email protected] 1Memorial University, 2Observatorio Nacional
3D potential field inversion for wireframe surface geometry (32 / 32)