Share
Embed
Transcript
180 slides
-
3D gravity inversion by planting anomalous densities Leonardo Uieda and
Valéria C. F. Barbosa August, 2011 Observatório Nacional -
Inverse Problem Planting Algorithm Synthetic Data Forward Problem Inspired by
René (1986) Outline -
Inverse Problem Planting Algorithm Synthetic Data Real Data Forward Problem
Inspired by René (1986) Outline -
Observations of g z Group in a vector: g= [g
1 g 2 ⋮ g N ] N×1 = observed data g Surface of the Earth -
= observed data g Assume caused by anomalous sources Δρ
Density contrast = -
Right rectangular prisms Parametrize the gravitational effect Discretize into M
elements Homogeneous density contrast jth element Linearize (Nagy et al., 2000) Interpretative model p j -
Arrange M density contrasts in a vector: Parametrize the gravitational
effect Discretize into M elements p= [p 1 p 2 ⋮ p M ] M×1 Parameter vector Linearize Interpretative model -
Discretize into M elements Parametrize the gravitational effect p j
=Δρ Prisms with not shown p j =0 -
Discretize into M elements Parametrize the gravitational effect g≈d Prisms
with not shown p j =0 p j =Δρ -
Discretize into M elements Parametrize the gravitational effect g≈d Predicted
data Prisms with not shown p j =0 p j =Δρ -
Discretize into M elements Parametrize the gravitational effect Gravitational effect
is linear d=∑ j=1 M p j a j g≈d Predicted data Prisms with not shown p j =0 p j =Δρ -
Discretize into M elements Parametrize the gravitational effect Gravitational effect
is linear d=∑ j=1 M p j a j Density contrast of jth prism g≈d Predicted data Prisms with not shown p j =0 p j =Δρ -
Discretize into M elements Parametrize the gravitational effect Gravitational effect
is linear d=∑ j=1 M p j a j g≈d Predicted data Effect of prism with unit density Prisms with not shown p j =0 p j =Δρ -
Discretize into M elements Parametrize the gravitational effect Gravitational effect
is linear g≈d Predicted data d=∑ j=1 M p j a j =A p Prisms with not shown p j =0 p j =Δρ -
Discretize into M elements Parametrize the gravitational effect Gravitational effect
is linear g≈d Predicted data Parameter vector d=∑ j=1 M p j a j =A p Prisms with not shown p j =0 p j =Δρ -
Discretize into M elements Parametrize the gravitational effect Gravitational effect
is linear g≈d Predicted data Jacobian (sensitivity) matrix d=∑ j=1 M p j a j =A p Prisms with not shown p j =0 p j =Δρ -
Discretize into M elements Parametrize the gravitational effect Gravitational effect
is linear g≈d Predicted data Column vector of A d=∑ j=1 M p j a j =A p Prisms with not shown p j =0 p j =Δρ -
Minimize difference between and g d r=g−d Residual vector Datamisfit
function: ϕ( p)=∥r∥2 = (∑ i=1 N (g i −d i )2 )1 2 -
Minimize difference between and g d r=g−d Residual vector Datamisfit
function: ϕ( p)=∥r∥2 = (∑ i=1 N (g i −d i )2 )1 2 ℓ2norm of r Leastsquares fit -
Constraints: 1. Compact no holes inside 2. Concentrated around “seeds”
Similar to René (1986) -
Constraints: 1. Compact no holes inside 2. Concentrated around “seeds”
• Userspecified prisms Similar to René (1986) -
Constraints: 1. Compact no holes inside 2. Concentrated around “seeds”
• Userspecified prisms • Given density contrasts ρs Similar to René (1986) -
Constraints: 1. Compact no holes inside 2. Concentrated around “seeds”
• Userspecified prisms • Given density contrasts • Any n° of ≠ density contrasts ρs Similar to René (1986) Not like René (1986) -
Constraints: 1. Compact no holes inside 2. Concentrated around “seeds”
• Userspecified prisms • Given density contrasts 3. Only • Any n° of ≠ density contrasts or p j =0 p j =ρs ρs Similar to René (1986) Not like René (1986) -
Constraints: 1. Compact no holes inside 2. Concentrated around “seeds”
• Userspecified prisms • Given density contrasts 3. Only • Any n° of ≠ density contrasts or p j =0 p j =ρs ρs 4. of closest seed p j =ρs Similar to René (1986) Not like René (1986) -
illposed problem wellposed problem constraints ϕ( p)=∥r∥2 = (∑ i=1
N (g i −d i )2 )1 2 Minimize data misfit Minimize goal function Γ( p)=ϕ( p)+μθ( p) -
illposed problem wellposed problem constraints ϕ( p)=∥r∥2 = (∑ i=1
N (g i −d i )2 )1 2 Minimize data misfit Minimize goal function Γ( p)=ϕ( p)+μθ( p) Regularizing parameter -
illposed problem wellposed problem constraints ϕ( p)=∥r∥2 = (∑ i=1
N (g i −d i )2 )1 2 Minimize data misfit Minimize goal function Γ( p)=ϕ( p)+μθ( p) Regularizing function -
illposed problem wellposed problem constraints ϕ( p)=∥r∥2 = (∑ i=1
N (g i −d i )2 )1 2 Minimize data misfit Minimize goal function Γ( p)=ϕ( p)+μθ( p) Regularizing function μ = tradeoff between fit and regularization -
Regularization: θ( p)=∑ j=1 M p j p j +ϵ
l j β Γ( p)=ϕ( p)+μθ( p) -
Regularization: θ( p)=∑ j=1 M p j p j +ϵ
l j β Γ( p)=ϕ( p)+μθ( p) Similar to Silva Dias et al. (2009) -
Regularization: θ( p)=∑ j=1 M p j p j +ϵ
l j β Γ( p)=ϕ( p)+μθ( p) ϵ = avoid singularity l j = distance between jth prism and seed β = how much compactness (3 to 7) Similar to Silva Dias et al. (2009) -
Regularization: θ( p)=∑ j=1 M p j p j +ϵ
l j β Γ( p)=ϕ( p)+μθ( p) For p j ≠0: -
Regularization: θ( p)=∑ j=1 M p j p j +ϵ
l j β Γ( p)=ϕ( p)+μθ( p) distance from seeds For p j ≠0: -
Regularization: θ( p)=∑ j=1 M p j p j +ϵ
l j β Γ( p)=ϕ( p)+μθ( p) distance from seeds regularizing function For p j ≠0: -
Regularization: θ( p)=∑ j=1 M p j p j +ϵ
l j β Γ( p)=ϕ( p)+μθ( p) distance from seeds regularizing function Imposes: • Compactness • Concentration around seeds For p j ≠0: -
Constraints: 1. Compact 2. Concentrated around “seeds” 3. Only or
p j =0 p j =Δρs 4. of closest seed p j =Δρs Regularization -
Constraints: 1. Compact 2. Concentrated around “seeds” 3. Only or
p j =0 p j =Δρs 4. of closest seed p j =Δρs Regularization Algorithm -
Based on René (1986) Start with seeds known density contrast
& position Overview: -
Based on René (1986) Start with seeds known density contrast
& position All other parameters set to 0 Overview: -
Based on René (1986) Start with seeds All other parameters
set to 0 Iteratively grow Overview: known density contrast & position -
Based on René (1986) Start with seeds All other parameters
set to 0 Iteratively grow add neighbor of seed Overview: known density contrast & position -
Based on René (1986) Start with seeds All other parameters
set to 0 Iteratively grow add neighbor of seed accretion Overview: known density contrast & position -
Based on René (1986) Start with seeds All other parameters
set to 0 Iteratively grow add neighbor of seed accretion Controlled by goal function and data misfit function Overview: Γ( p)=ϕ( p)+μθ( p) ϕ( p)=∥r∥2 = (∑ i=1 N (g i −d i )2 )1 2 known density contrast & position -
Algorithm: Define interpretative model All parameters zero g = observed
data Interpretative model -
Algorithm: seeds N S Define interpretative model All parameters zero
g = observed data Interpretative model -
Algorithm: seeds N S Define interpretative model All parameters zero
Include seeds Prisms with not shown p j =0 Seeds -
Algorithm: seeds N S Define interpretative model All parameters zero
Include seeds Compute initial residuals r(0)=g−d(0) Prisms with not shown p j =0 -
Algorithm: Residual vector seeds N S Define interpretative model All
parameters zero Include seeds Compute initial residuals r(0)=g−d(0) Prisms with not shown p j =0 -
Algorithm: Observed data seeds N S Define interpretative model All
parameters zero Include seeds Compute initial residuals r(0)=g−d(0) g = observed data Prisms with not shown p j =0 -
Algorithm: Predicted by seeds seeds N S Define interpretative model
All parameters zero Include seeds Compute initial residuals r(0)=g−d(0) g = observed data d = predicted data Prisms with not shown p j =0 -
Algorithm: seeds N S Define interpretative model All parameters zero
Include seeds Compute initial residuals r(0)=g−d(0) g = observed data d=∑ j=0 M p j a j d = predicted data Prisms with not shown p j =0 -
Algorithm: seeds N S Define interpretative model All parameters zero
Include seeds Compute initial residuals r(0)=g−d(0) g = observed data d=∑ j=0 M p j a j Many=0 d = predicted data Prisms with not shown p j =0 -
Algorithm: seeds N S Define interpretative model All parameters zero
Include seeds Compute initial residuals r(0)=g−d(0) g = observed data d=∑ s=1 N S ρ s a j S d = predicted data Prisms with not shown p j =0 -
Algorithm: seeds N S Define interpretative model All parameters zero
Include seeds Compute initial residuals r(0)=g− (∑ s=1 N S ρ s a j S ) Prisms with not shown g = observed data d = predicted data p j =0 -
Algorithm: Density contrast of sth seed seeds N S Define
interpretative model All parameters zero Include seeds Compute initial residuals r(0)=g− (∑ s=1 N S ρ s a j S ) Prisms with not shown g = observed data d = predicted data p j =0 -
seeds N S Algorithm: Define interpretative model All parameters zero
Include seeds Compute initial residuals r(0)=g− (∑ s=1 N S ρ s a j S ) Column vector of A Prisms with not shown g = observed data d = predicted data p j =0 -
seeds N S Algorithm: Define interpretative model All parameters zero
Include seeds Compute initial residuals r(0)=g− (∑ s=1 N S ρ s a j S ) Prisms with not shown g = observed data d = predicted data Neighbors Find neighbors of seeds p j =0 -
Prisms with not shown Growth: Try accretion to sth seed:
Choose neighbor: p j =0 -
Prisms with not shown Growth: Try accretion to sth seed:
Choose neighbor: 1. Reduce data misfit ϕ( p)=∥r∥ 2 p j =0 -
Prisms with not shown Growth: Try accretion to sth seed:
Choose neighbor: 1. Reduce data misfit 2. Smallest goal function ϕ( p)=∥r∥ 2 Γ( p)=ϕ( p)+μθ( p) p j =0 -
Prisms with not shown Growth: Try accretion to sth seed:
Choose neighbor: 1. Reduce data misfit 2. Smallest goal function ϕ( p)=∥r∥ 2 Γ( p)=ϕ( p)+μθ( p) j = chosen j p j =0 -
Prisms with not shown Growth: Try accretion to sth seed:
Choose neighbor: 1. Reduce data misfit 2. Smallest goal function ϕ( p)=∥r∥ 2 Γ( p)=ϕ( p)+μθ( p) p j =ρ s j = chosen j p j =0 (New elements) -
Prisms with not shown Growth: Try accretion to sth seed:
Choose neighbor: 1. Reduce data misfit 2. Smallest goal function ϕ( p)=∥r∥ 2 Γ( p)=ϕ( p)+μθ( p) p j =ρ s j = chosen j p j =0 (New elements) new predicted data -
Prisms with not shown Growth: Try accretion to sth seed:
Choose neighbor: 1. Reduce data misfit 2. Smallest goal function ϕ( p)=∥r∥ 2 Γ( p)=ϕ( p)+μθ( p) p j =ρ s j = chosen Update residuals r(new)=g−d(new) p j =0 (New elements) new predicted data j -
Prisms with not shown Growth: Try accretion to sth seed:
Choose neighbor: 1. Reduce data misfit 2. Smallest goal function ϕ( p)=∥r∥ 2 Γ( p)=ϕ( p)+μθ( p) p j =ρ s j = chosen Update residuals r(new)=g−d(new) p j =0 (New elements) new predicted data j d(old)+ effect of j -
Prisms with not shown Growth: Try accretion to sth seed:
Choose neighbor: 1. Reduce data misfit 2. Smallest goal function ϕ( p)=∥r∥ 2 Γ( p)=ϕ( p)+μθ( p) p j =ρ s j = chosen Update residuals r(new)=g−d(new) p j =0 (New elements) new predicted data j d(old)+ effect of j ∑ s=1 N S ρs a j S p j a j + -
Prisms with not shown Growth: Try accretion to sth seed:
Choose neighbor: 1. Reduce data misfit 2. Smallest goal function ϕ( p)=∥r∥ 2 Γ( p)=ϕ( p)+μθ( p) p j =ρ s j = chosen Update residuals r(new)=g−d(new) p j =0 (New elements) new predicted data j d(old)+ effect of j ∑ s=1 N S ρs a j S p j a j + -
Prisms with not shown Growth: Try accretion to sth seed:
Choose neighbor: 1. Reduce data misfit 2. Smallest goal function ϕ( p)=∥r∥ 2 Γ( p)=ϕ( p)+μθ( p) p j =ρ s j = chosen Update residuals r(new)=g−∑ s=1 N S ρs a j S − p j a j p j =0 (New elements) new predicted data j -
Prisms with not shown Growth: Try accretion to sth seed:
Choose neighbor: 1. Reduce data misfit 2. Smallest goal function ϕ( p)=∥r∥ 2 Γ( p)=ϕ( p)+μθ( p) p j =ρ s j = chosen Update residuals r(new)=g−∑ s=1 N S ρs a j S − p j a j p j =0 (New elements) new predicted data j { r(0) -
Prisms with not shown Growth: Try accretion to sth seed:
Choose neighbor: 1. Reduce data misfit 2. Smallest goal function ϕ( p)=∥r∥ 2 Γ( p)=ϕ( p)+μθ( p) p j =ρ s j = chosen Update residuals p j =0 (New elements) new predicted data j r(new)=r(old )− p j a j -
Prisms with not shown Growth: None found = no accretion
Try accretion to sth seed: 1. Reduce data misfit 2. Smallest goal function p j =ρ s j = chosen Update residuals r(new)=r(old )− p j a j Choose neighbor: p j =0 -
Prisms with not shown Growth: None found = no accretion
Try accretion to sth seed: 1. Reduce data misfit 2. Smallest goal function p j =ρ s j = chosen Update residuals r(new)=r(old )− p j a j Choose neighbor: Variable sizes p j =0 -
Prisms with not shown Growth: None found = no accretion
N S Try accretion to sth seed: 1. Reduce data misfit 2. Smallest goal function p j =ρ s j = chosen Update residuals r(new)=r(old )− p j a j Choose neighbor: p j =0 -
Prisms with not shown Growth: None found = no accretion
N S Try accretion to sth seed: 1. Reduce data misfit 2. Smallest goal function p j =ρ s j = chosen Update residuals r(new)=r(old )− p j a j Choose neighbor: p j =0 (New elements) j -
Prisms with not shown Growth: None found = no accretion
N S Try accretion to sth seed: 1. Reduce data misfit 2. Smallest goal function p j =ρ s j = chosen Update residuals r(new)=r(old )− p j a j Choose neighbor: At least one seed grow? Yes No p j =0 -
Prisms with not shown Growth: None found = no accretion
N S Try accretion to sth seed: 1. Reduce data misfit 2. Smallest goal function p j =ρ s j = chosen Update residuals r(new)=r(old )− p j a j Choose neighbor: At least one seed grow? Yes No p j =0 -
Prisms with not shown Growth: None found = no accretion
N S Try accretion to sth seed: 1. Reduce data misfit 2. Smallest goal function p j =ρ s j = chosen Update residuals r(new)=r(old )− p j a j Choose neighbor: At least one seed grow? Yes No p j =0 (New elements) j -
Prisms with not shown Growth: None found = no accretion
N S Try accretion to sth seed: 1. Reduce data misfit 2. Smallest goal function p j =ρ s j = chosen Update residuals r(new)=r(old )− p j a j Choose neighbor: At least one seed grow? Yes No p j =0 (New elements) j -
Prisms with not shown Growth: None found = no accretion
N S Try accretion to sth seed: 1. Reduce data misfit 2. Smallest goal function p j =ρ s j = chosen Update residuals r(new)=r(old )− p j a j Choose neighbor: At least one seed grow? Yes No p j =0 j -
Prisms with not shown Growth: None found = no accretion
N S Try accretion to sth seed: 1. Reduce data misfit 2. Smallest goal function p j =ρ s j = chosen Update residuals r(new)=r(old )− p j a j Choose neighbor: At least one seed grow? Yes No p j =0 j -
Prisms with not shown Growth: None found = no accretion
N S Try accretion to sth seed: 1. Reduce data misfit 2. Smallest goal function p j =ρ s j = chosen Update residuals r(new)=r(old )− p j a j Choose neighbor: At least one seed grow? Yes No p j =0 j -
Prisms with not shown Growth: None found = no accretion
N S Try accretion to sth seed: 1. Reduce data misfit 2. Smallest goal function p j =ρ s j = chosen Update residuals r(new)=r(old )− p j a j Choose neighbor: At least one seed grow? Yes No p j =0 j -
Prisms with not shown Growth: None found = no accretion
N S Try accretion to sth seed: 1. Reduce data misfit 2. Smallest goal function p j =ρ s j = chosen Update residuals r(new)=r(old )− p j a j Choose neighbor: At least one seed grow? Yes No p j =0 j -
Prisms with not shown Growth: None found = no accretion
N S Try accretion to sth seed: 1. Reduce data misfit 2. Smallest goal function p j =ρ s j = chosen Update residuals r(new)=r(old )− p j a j Choose neighbor: At least one seed grow? Yes No p j =0 j -
Prisms with not shown Growth: None found = no accretion
N S Try accretion to sth seed: 1. Reduce data misfit 2. Smallest goal function p j =ρ s j = chosen Update residuals r(new)=r(old )− p j a j Choose neighbor: At least one seed grow? Yes No p j =0 -
Prisms with not shown Growth: None found = no accretion
N S Try accretion to sth seed: 1. Reduce data misfit 2. Smallest goal function p j =ρ s j = chosen Update residuals r(new)=r(old )− p j a j Choose neighbor: At least one seed grow? Yes No Done! p j =0 -
Advantages: Compact & nonsmooth Any number of sources Any number
of different density contrasts No large equation system Search limited to neighbors -
Remember equations: r(0)=g− (∑ s=1 N S ρs a j
S ) r(new)=r(old)− p j a j Initial residual Update residual vector -
Remember equations: r(0)=g− (∑ s=1 N S ρ s a
j S ) r(new)=r(old)− p j a j Initial residual Update residual vector No matrix multiplication (only vector +) -
No matrix multiplication (only vector +) Remember equations: r(0)=g− (∑
s=1 N S ρ s a j S ) r(new)=r(old)− p j a j Initial residual Update residual vector Only need some columns of A -
No matrix multiplication (only vector +) Remember equations: r(0)=g− (∑
s=1 N S ρ s a j S ) r(new)=r(old)− p j a j Initial residual Update residual vector Only need some columns of A Calculate only when needed -
No matrix multiplication (only vector +) Remember equations: r(0)=g− (∑
s=1 N S ρ s a j S ) r(new)=r(old)− p j a j Initial residual Update residual vector Only need some columns of A Calculate only when needed & delete after update -
No matrix multiplication (only vector +) Remember equations: r(0)=g− (∑
s=1 N S ρ s a j S ) r(new)=r(old)− p j a j Initial residual Update residual vector Only need some columns of A Calculate only when needed Lazy evaluation & delete after update -
Advantages: Compact & nonsmooth Any number of sources Any number
of different density contrasts No large equation system Search limited to neighbors -
Advantages: Compact & nonsmooth Any number of sources Any number
of different density contrasts No large equation system Search limited to neighbors No matrix multiplication (only vector +) Lazy evaluation of Jacobian -
Advantages: Compact & nonsmooth Any number of sources Any number
of different density contrasts No large equation system Search limited to neighbors No matrix multiplication (only vector +) Lazy evaluation of Jacobian Fast inversion + low memory usage -
Δρ=1.0 g/cm3 Δρ=0.5 g/cm3 • Sources = 1 km X
1 km X 1 km -
• Sources = 1 km X 1 km X 1
km Depth=0.8 km -
• Sources = 1 km X 1 km X 1
km Depth=1.6 km Depth=0.8 km -
• Sources = 1 km X 1 km X 1
km • Data set = 375 observations Depth=1.6 km Depth=0.8 km -
• Sources = 1 km X 1 km X 1
km • Area = 5 km X 3 km • Data set = 375 observations Depth=1.6 km Depth=0.8 km -
• 0.05 mGal Gaussian noise • Sources = 1 km
X 1 km X 1 km • Area = 5 km X 3 km • Data set = 375 observations Depth=1.6 km Depth=0.8 km -
• Interpretative model = 151,875 prisms • Prisms = 66.7
m X 66.7 m X 66.7 m -
• Used 2 seeds • With corresponding density contrasts •
Placed in center of sources -
Δρ=0.5 g/cm3 • Used 2 seeds • With corresponding density
contrasts • Placed in center of sources -
Δρ=1.0 g/cm3 Δρ=0.5 g/cm3 • Used 2 seeds • With
corresponding density contrasts • Placed in center of sources -
Inversion result • compact • concentrated around seeds • recover
correct geometry of sources -
Predicted data Inversion result • compact • concentrated around seeds
• recover correct geometry of sources -
Predicted data Observed data Inversion result • compact • concentrated
around seeds • recover correct geometry of sources -
Predicted data Observed data Inversion result • compact • concentrated
around seeds • fits observations • recover correct geometry of sources -
Predicted data Observed data On laptop with 2.0 GHz •
375 data • 151,875 prisms • Total time≈4.4 min -
Cana Brava complex (CBC) & Palmeirópolis sequence (PVSS) • Outcropping
• North of Goiás • Tocantins Province • Amazonian & São Francisco cratons After Carminatti et al. (2003) -
Cana Brava complex (CBC) & Palmeirópolis sequence (PVSS) Gravimetric data:
• 132 observations • Residual Bouguer • Max 45 mGal After Carminatti et al. (2003) -
Cana Brava complex (CBC) & Palmeirópolis sequence (PVSS) Previous interpretation:
After Carminatti et al. (2003) • Carminatti et al. (2003) • PVSS • CVC • Max depth Δρ=0.27 g/cm3 Δρ=0.39g/cm3 ≈6 km -
Cana Brava complex (CBC) & Palmeirópolis sequence (PVSS) Previous interpretation:
After Carminatti et al. (2003) • Carminatti et al. (2003) • PVSS • CVC • Max depth Δρ=0.27 g/cm3 Δρ=0.39g/cm3 ≈6 km Test this hypothesis -
Assign seeds Green: z=0 km Δρ=0.27g/cm3 Blue: z=2 km Δρ=0.27g/cm3
Red: z=0 km Δρ=0.39 g/cm3 Total = 269 Assign seeds -
Interpretative model Size: 120 km X 50 km X 11
km 480,000 prisms Prism size: 500 m X 500 m X 575 m -
Inversion result On laptop with 2.0 GHz • 132 data
• 480,00 prisms • Total time≈3.75 min -
• New 3D gravity inversion • Multiple sources • Interfering
gravitational effects • Abrupt densitycontrast distribution • No matrix multiplication • No need to solve large linear systems • Ideal for: ore bodies, intrusions, salt domes, etc Conclusions -
• Developed for gravity gradients • Presented at EAGE 2011
preliminary results • To be presented at SEG 2011: • Final results • Robust method to handle nontargeted sources Previous and future work