3D gravity inversion incorporating prior information through an adaptive learning procedure Fernando J. S. Silva Dias Valéria C. F. Barbosa National Observatory.

Slides:



Advertisements
Similar presentations
Applications of one-class classification
Advertisements

Problem 1 The corrections can be larger than the anomaly Stat.Time T Dist. (m) Elev. (m) Reading (dial units) Base reading at time T Drift corr’d anom.
Point-wise Discretization Errors in Boundary Element Method for Elasticity Problem Bart F. Zalewski Case Western Reserve University Robert L. Mullen Case.
Fast Bayesian Matching Pursuit Presenter: Changchun Zhang ECE / CMR Tennessee Technological University November 12, 2010 Reading Group (Authors: Philip.
Aspects of Conditional Simulation and estimation of hydraulic conductivity in coastal aquifers" Luit Jan Slooten.
Nice, 17/18 December 2001 Adaptive Grids For Bathymetry Mapping And Navigation Michel Chedid and Maria-João Rendas I3S - MAUVE.
Paper Discussion: “Simultaneous Localization and Environmental Mapping with a Sensor Network”, Marinakis et. al. ICRA 2011.
Prénom Nom Document Analysis: Linear Discrimination Prof. Rolf Ingold, University of Fribourg Master course, spring semester 2008.
L15:Microarray analysis (Classification) The Biological Problem Two conditions that need to be differentiated, (Have different treatments). EX: ALL (Acute.
Classification with reject option in gene expression data Blaise Hanczar and Edward R Dougherty BIOINFORMATICS Vol. 24 no , pages
Real-time Combined 2D+3D Active Appearance Models Jing Xiao, Simon Baker,Iain Matthew, and Takeo Kanade CVPR 2004 Presented by Pat Chan 23/11/2004.
Geostatistical structural analysis of TransCom data for development of time-dependent inversion Erwan Gloaguen, Emanuel Gloor, Jorge Sarmiento and TransCom.
Moments area of the object center of mass  describe the image content (or distribution) with respect to its axes.
Gravity: Gravity anomalies. Earth gravitational field. Isostasy. Moment density dipole. Practical issues.
Advanced Topics in Optimization
Parallel K-Means Clustering Based on MapReduce The Key Laboratory of Intelligent Information Processing, Chinese Academy of Sciences Weizhong Zhao, Huifang.
Robust Statistical Estimation of Curvature on Discretized Surfaces Evangelos Kalogerakis Patricio Simari Derek Nowrouzezahrai Karan Singh Symposium on.
Motion from normal flow. Optical flow difficulties The aperture problemDepth discontinuities.
SINGLE CHANNEL SPEECH MUSIC SEPARATION USING NONNEGATIVE MATRIXFACTORIZATION AND SPECTRAL MASKS Jain-De,Lee Emad M. GraisHakan Erdogan 17 th International.
The aim of the presented research activities Is to develop new interpretation techniques for potential fields exploration methods (gravity, magnetic,
Gravity I: Gravity anomalies. Earth gravitational field. Isostasy.
1 Preview At least two views are required to access the depth of a scene point and in turn to reconstruct scene structure Multiple views can be obtained.
Radial gravity inversion constrained by total anomalous mass excess for retrieving 3D bodies Vanderlei Coelho Oliveira Junior Valéria C. F. Barbosa Observatório.
Linear(-ized) Inverse Problems
Course 12 Calibration. 1.Introduction In theoretic discussions, we have assumed: Camera is located at the origin of coordinate system of scene.
An Autonomic Framework in Cloud Environment Jiedan Zhu Advisor: Prof. Gagan Agrawal.
Polynomial Equivalent Layer Valéria C. F. Barbosa* Vanderlei C. Oliveira Jr Observatório Nacional.
Chapter 21 R(x) Algorithm a) Anomaly Detection b) Matched Filter.
Adaptive learning gravity inversion for 3D salt body imaging Fernando J. S. Silva Dias Valéria C. F. Barbosa National Observatory João B. C. Silva Federal.
Bayesian Inversion of Stokes Profiles A.Asensio Ramos (IAC) M. J. Martínez González (LERMA) J. A. Rubiño Martín (IAC) Beaulieu Workshop ( Beaulieu sur.
Graphical Separation of Residual
Ch 2. Probability Distributions (1/2) Pattern Recognition and Machine Learning, C. M. Bishop, Summarized by Yung-Kyun Noh and Joo-kyung Kim Biointelligence.
Testing of the harmonic inversion method on the territory of the eastern part of Slovakia.
Linear Programming Erasmus Mobility Program (24Apr2012) Pollack Mihály Engineering Faculty (PMMK) University of Pécs João Miranda
A Flexible New Technique for Camera Calibration Zhengyou Zhang Sung Huh CSPS 643 Individual Presentation 1 February 25,
112/5/ :54 Graphics II Image Based Rendering Session 11.
Boosted Particle Filter: Multitarget Detection and Tracking Fayin Li.
Multiple Instance Learning for Sparse Positive Bags Razvan C. Bunescu Machine Learning Group Department of Computer Sciences University of Texas at Austin.
On Optimization Techniques for the One-Dimensional Seismic Problem M. Argaez¹ J. Gomez¹ J. Islas¹ V. Kreinovich³ C. Quintero ¹ L. Salayandia³ M.C. Villamarin¹.
Large-Scale Matrix Factorization with Missing Data under Additional Constraints Kaushik Mitra University of Maryland, College Park, MD Sameer Sheoreyy.
A table of diagnostic positions and depth index multipliers for the Sphere (see your handout). Note that regardless of which diagnostic position you use,
CSCE 641 Computer Graphics: Image-based Rendering (cont.) Jinxiang Chai.
Zhilin Zhang, Bhaskar D. Rao University of California, San Diego March 28,
Interactive 2D magnetic inversion: a tool for aiding forward modeling and testing geological hypotheses Valéria C. F. Barbosa LNCC - National Laboratory.
STATIC ANALYSIS OF UNCERTAIN STRUCTURES USING INTERVAL EIGENVALUE DECOMPOSITION Mehdi Modares Tufts University Robert L. Mullen Case Western Reserve University.
Inverse Modeling of Surface Carbon Fluxes Please read Peters et al (2007) and Explore the CarbonTracker website.
3D depth-to-basement and density contrast estimates using gravity and borehole data Cristiano Mendes Martins Valéria C. F. Barbosa National Observatory.
Ch 2. Probability Distributions (1/2) Pattern Recognition and Machine Learning, C. M. Bishop, Summarized by Joo-kyung Kim Biointelligence Laboratory,
Media Processor Lab. Media Processor Lab. Trellis-based Parallel Stereo Matching Media Processor Lab. Sejong univ.
1 Introduction Optimization: Produce best quality of life with the available resources Engineering design optimization: Find the best system that satisfies.
Dario Grana and Tapan Mukerji Sequential approach to Bayesian linear inverse problems in reservoir modeling using Gaussian mixture models SCRF Annual Meeting,
Alexandra Moshou, Panayotis Papadimitriou and Kostas Makropoulos MOMENT TENSOR DETERMINATION USING A NEW WAVEFORM INVERSION TECHNIQUE Department of Geophysics.
1 Double-Patterning Aware DSA Template Guided Cut Redistribution for Advanced 1-D Gridded Designs Zhi-Wen Lin and Yao-Wen Chang National Taiwan University.
Presented by 翁丞世  View Interpolation  Layered Depth Images  Light Fields and Lumigraphs  Environment Mattes  Video-Based.
Fast Least Squares Migration with a Deblurring Filter 30 October 2008 Naoshi Aoki 1.
Linear Discriminant Functions Chapter 5 (Duda et al.) CS479/679 Pattern Recognition Dr. George Bebis.
Biointelligence Laboratory, Seoul National University
3D structure of the Thuringian Basin, Germany
4D Gravity Inversion Hyoungrea Bernard Rim
On Optimization Techniques for the One-Dimensional Seismic Problem
Two-view geometry Computer Vision Spring 2018, Lecture 10
Centroids Lesson 7.5.
Unfolding Problem: A Machine Learning Approach
Assessing uncertainties on production forecasting based on production Profile reconstruction from a few Dynamic simulations Gaétan Bardy – PhD Student.
Combining Geometric- and View-Based Approaches for Articulated Pose Estimation David Demirdjian MIT Computer Science and Artificial Intelligence Laboratory.
Outline Derivatives and transforms of potential fields
Figure 1.1 The parabolic trajectory problem.
Biointelligence Laboratory, Seoul National University
Unfolding with system identification
Advanced deconvolution techniques and medical radiography
Presentation transcript:

3D gravity inversion incorporating prior information through an adaptive learning procedure Fernando J. S. Silva Dias Valéria C. F. Barbosa National Observatory João B. C. Silva Federal University of Pará

Content Introduction and Objective Methodology Real Data Inversion Result Conclusions Synthetic Data Inversion Results

To estimate 3D source geometries that may give rise to interfering gravity anomalies. Introduction and Objective Objective Methods that estimate 3D density-contrast distributions: Bear et al. (1995) Li and Oldenburg (1998) Portniaguine and Zhdanov (1999) Zhdanov et al. (2004) density contrast (g/cm 3 )

Methodology Forward modeling of gravity anomalies Inverse Problem Adaptive Learning Procedure

Gravity anomaly x y z 3D gravity sources Source Region Forward modeling of gravity anomalies y x Depth

y z x Source Region dy dz dx The source region is divided into an mx × my × mz grid of M 3D vertical juxtaposed prisms Forward modeling of gravity anomalies

x Observed gravity anomaly y z Depth Source Region To estimate the 3D density-contrast distribution y x Forward modeling of gravity anomalies

The vertical component of the gravity field produced by the density-contrast distribution  ( r’ ): )(g i r )'(r  V i ' ' ' 3   i dv zz rr   Methodology The discrete forward modeling operator for the gravity anomaly can be expressed by: g  A p ' ' ' )( 3     j V i i iij dv zz A rr r  where Steiner (1978) (N x 1)(M x 1)(NxM)

Methodology 2 A g o  1  N g  The unconstrained Inverse Problem The linear inverse problem can be formulated by minimizing The problem of obtaining a vector of parameter estimates, p, that minimizes this functional is an ill-posed problem. ^ p

x y z Source Region Depth Methodology Concentration of mass excess about N E specified geometric elements (axes and points)

Methodology Iterative inversion method that: The density-contrast distribution must assume just two known density contrast values: zero The estimated nonnull density contrast must be concentrated about a set of N E geometric elements (axes and points)  fits the gravity data  satisfies two constraints: or a nonnull value.

The method estimates iteratively the constrained parameter correction Δp by Minimizing Subject to Methodology Δp 2 )( k W )( k 1/2 p and updates the density-contrast estimates by 2 A g o  1 N   Δp ) (p o + )( k )( k )( )( )1( ˆˆ k k k pΔpp o    ≡ )( 3 ˆ k-1 j j jj p d w WpWp )( k 1/2 )( k = {} Prior reference vector

} { min 1N j d E d j     MjNzezyeyxd E jjj,,1,,,1)()( ( 2/1 222  xe) j   Methodology z y x xe ) ye,, ze ) j d The method defines d j as the distance from the center of the j th prism to the closest geometric element closest geometric element d j

M jdp target j,...,1},{min arg, j * *  E n1   }{ min 1 j N j dd E   z y x xe ) ye,, ze ) Methodology The method assigns to the j th prism the target density contrast of the geometric element closest to the j th prism d j

Methodology z x axis point j y j d  g/cm 3  g/cm 3  g/cm 3 At the first iteration: Initial interpretation model First-guess geometric elements The corresponding target density contrasts Static Geologic Reference Model p j target  g/cm 3 j d and j p target The method assigns to each prism a pair of:.

Methodology Penalization Algorithm: )( ˆ k j p j p target 0 (g/cm 3 ) j p target 0 (g/cm 3 ) For positive target density contrast For negative target density contrast )( ˆ k j p )( ˆ k j p )( ˆ k j p jj wpwp )( k 1/2 =   target j p or 0 (g/cm 3 ) )( ˆ k pΔ )(k p o  )1( ˆ k p  ( k ) o p j 

Methodology Penalization Algorithm: j p target 0 (g/cm 3 ) j p target 0 (g/cm 3 ) For positive target density contrast For negative target density contrast )( ˆ k j p )( )( )1( ˆˆ k k k pΔpp o   p j target 2 p j 2 )( ˆ k j p )( ˆ k j p )( ˆ k j p j p ( k ) o p j  o p j  0 (g/cm 3 )  )( 3 ˆ k-1 j j jj p d wpwp )( k 1/2 = )( ˆ k j p )( ˆ k j p

The choice of the interpretation model

noise-corrupted gravity anomaly geometric element The choice of the interpretation model True source  target = 0.4 g/cm 3.  = 0.4 g/cm 3.

True source Fitted anomaly Rough interpretation model: 4×4×4 grid of 3D prisms First x(km) y(km) True source density contrast (g/cm 3 ) Rough interpretation model : 5×5×5 grid of 3D prisms Fitted anomaly Second

True source Refined interpretation model: 24×24×24 grid of 3D prisms Fourth density contrast (g/cm 3 ) Fitted anomaly True source Refined interpretation model: 12×12×12 grid of 3D prisms Third Fitted anomaly

Adaptive Learning Procedure New interpretation model New geometric elements Associated target density contrasts

Adaptive Learning Procedure x y z Source Region First Iteration

Adaptive Learning Procedure x y z Each 3D prism is divided Second Iteration

Adaptive Learning Procedure x y z Iteration  Iteration  Iteration  Iteration  New interpretation model

static geologic reference model x y z First Iteration Second Iteration New geometric elements (points) and associated target density contrasts Dynamic geologic reference model Adaptive Learning Procedure First interpretation model and the static geologic reference model First density-contrast distribution estimate New interpretation model

Adaptive Learning Procedure Static geologic reference model  target = 0.4 g/cm 3.

Adaptive Learning Procedure Fourth iteration density contrast (g/cm 3 ) 0.40 Without using the adaptive learning procedure Both interpretation models consist of 24×24×24 grid of 3D prisms density contrast (g/cm 3 ) 0.40 True source

Inversions of Synthetic Data

Large source surrounding a small source y (km) x (km) density contrast (g/cm 3 ) Anorthosite ( 0.4 g/m 3 ) Granite ( 0.2 g/cm 3 )

Large source surrounding a small source The red dots are the first-guess skeletal outlines: static geologic reference model

Large source surrounding a small source First iteration Interpretation model: 4×3×3 grid of 3D prisms. density contrast (g/cm 3 ) Fitted anomaly

Large source surrounding a small source Fourth iteration interpretation model : 32×24×24 grid of 3D prisms. Fitted anomaly density contrast (g/cm 3 )

Multiple buried sources at different depths 0.15 g/cm 3 0.3g/cm g/cm 3 The axes are the first-guess skeletal outlines: static geologic reference model density contrast (g/cm 3 )

Multiple buried sources at different depths Third iteration Interpretation model: 28×48×24 grid of 3D prisms. density contrast (g/cm 3 ) Fitted anomaly

Real Gravity Data Redenção Granite (Brazil)

The Amazon Craton in northern Brazil, within the Archean Greenstone unit, comprising a part of the Carajás metallogenic province. Localization and Geological Setting Oliveira et al. (2007) Brazil

Geologic Map of the Redenção Area SRTM / Gamma Thorium Oliveira et al. (2007)

The red dots are the first-guess skeletal outlines static reference model Redenção Granite The associated target density contrasts are: g/cm 3 or g/cm g/cm g/cm 3

Redenção Granite density contrast (g/cm 3 ) Fitted anomaly Fourth iteration Interpretation model : 64×72×32 grid of 3D prisms. Dynamic Geometric Elements

Conclusions

3D gravity inversion incorporating prior information through an adaptive learning procedure  Estimates 3D source geometries that may give rise to an interfering gravity anomaly  Concentrates the largest density contrast estimates about first-guess skeletal outlines  Creates new skeletal outlines and a new refined interpretation model  Makes it possible to reconstruct a sharp image of multiple and closely located gravity sources. The proposed method: density contrast (g/cm 3 ) first-guess skeletal outlines density contrast (g/cm 3 ) New skeletal

Thank You I cordially invite you to attend the upcoming

Extra Figures 1 CPU ATHLON with one core and 2.4 GHertz and 1 MB of cache L2 2GB of DDR1 memory

 = 0.4 g/cm 3. Isolated gravity anomalies

density contrast (g/cm 3 ) Li and Oldenburg (1998)

density contrast (g/cm 3 ) Portniaguine and Zhdanov (2002)

density contrast (g/cm 3 ) Our gravity inversion

Interfering gravity anomalies

density contrast (g/cm 3 ) Li and Oldenburg (1998)

density contrast (g/cm 3 ) Portniaguine and Zhdanov (2002)

density contrast (g/cm 3 ) Our gravity inversion

Methodology The Iterative Constrained Inverse Problem Starting from the minimum-norm solution o gIAAA p 1o ) ( ˆ   TT we look, at the kth iteration, for a constrained parameter correction ˆˆ ˆ )( o pΔp p   ) 1 ( k  k )( k and update the density-contrast distribution estimate by  )(11)( ˆ )( ˆ k o T )(k T )(k k pA g IAA WAWpΔ o  1    ,