Dimensionality Reduction for Seismic Attribute Analysis Bradley C. - PowerPoint PPT Presentation

Dimensionality Reduction for Seismic Attribute Analysis Bradley C. Wallet, Ph.D. University of Oklahoma ConocoPhillips School of Geology and Geophysics

Where oil is first found is in the minds’ of men - Wallace Pratt

Motivation

Motivation

Outline • Seismic data • Seismic attributes • PCA • Image grand tour • Non-linear methods • Conclusions • Acknowledgements

Outline • Seismic data • Seismic attributes • PCA • Image grand tour • Non-linear methods • Conclusions • Acknowledgements

Why care about seismic data? • Single pre-stack data sets can be 10’s – 100’s of terabytes in size • Provide good spatial coverage exploration area • Used to make high dollar decisions

Seismic shot Courtesy of Bin Lyu

Common midpoint gather

Migration

Convolutional model Reflection Lithology Velocity Density Impedance Wavelet Coefficients Shale ⇒ ⇒ Sand x = * Shale Sand Shale

Seismic data (Elebiju et al., 2009)

Outline • Seismic data • Seismic attributes • PCA • Image grand tour • Non-linear methods • Conclusions • Acknowledgements

These are features

From one comes many Seismic data Attribute 1 Attribute 2 Attribute 3 Attribute 4 Attribute 5 Attribute 6 Attribute 7 Attribute 8

Coherence inline inline

Seismic 5 km (Bahorich and Farmer, 1995)

Coherence Coh 1.0 0.6 5 km salt (Bahorich and Farmer, 1995)

Σ Spectral decomposition Reflectivity Synthetic CWT Magnitude Voices CWT magnitude pos 0 Le Nozze di Figaro (Matos and Marfurt, 2011)

Spectral decomposition A ′ A Time (s) 30 Hz 15 Hz A A A ′ A ′ 30 Hz Map (Laughlin et al., 2002) 15 Hz Map

Spectral decomposition 18 Hz  Red 24 Hz  Green 36 Hz  Blue (Bahorich et al., 2002)

Dip attributes z θ (dip magnitude) φ (dip azimuth) n a θ y θ x (crossline dip) (inline dip) ψ y x (strike) (Marfurt, 2006)

Dip attributes Minimum dip tested (-20 0 ) Dip with maximum coherence (+5 0 ) Analysis Point Maximum dip tested (+20 0 ) Instantaneous dip = dip with highest coherence (Marfurt et al, 1998)

Dip attributes Dip Azimuth Hue 0 180 360 High Dip Magnitude Saturation 0 N 1.2 E W 1.4 S (c) (Guo et al., 2008)

How do we “assimilate” all these attributes?

Outline • Seismic data • Seismic attributes • PCA • Image grand tour • Non-linear methods • Conclusions • Acknowledgements

PCA • Rotates attribute space • New dimensions are called principal components • Var(pc1) > Var(pc2) > … > Var(pc d) • Defines variance as information

PCA (Wikapedia)

Watonga survey

Complex PCA

Complex PCA

PCA

PCA

PCA

Outline • Seismic data • Seismic attributes • PCA • Image grand tour • Non-linear methods • Conclusions • Acknowledgements

Linear projections Poorly separated Well separated Somewhat separated ξ d ∑ = α ξ proj ( ) i i = i 1

The Grand Tour (1750-1880’s)

Defining the tour

Image Grand Tour 7.005 -6.215 10.95

View Locked Color IGT

Outline • Seismic data • Seismic attributes • PCA • Image grand tour • Non-linear methods • Conclusions • Acknowledgements

Latent spaces

Generative topographical maps a) N b) Cartoon illustration of GTM

Waka 3D Canterbury Basin, offshore New Zealand 170° 30’ E 173° 00’ E 45° 30’ S 46° 30’ S (Modified from Mitchell and Neil, 2012) (Figure by Origin Energy)

Seismic 36

Peak Frequency

Peak spectral magnitude 38

Curvedness 39

GLCM homogeneity 40

Co-rendering 41

GTM

Waveforms as attributes (Wallet et al, 2009)

Watonga revisited (Wallet et al, 2009)

Diffusion maps Form n-by-n similarity matrix Normalize rows to sum to 1 Perform PCA on diffusion matrix

Diffusion maps Advantages Disadvantages • Closed form solution • Computationally intractable for reasonable sized data sets • Direct calculation of inter-point distances • Out of training set data are not defined in mapping • Not tied to a Euclidean space • Eigenvalues

Diffusion maps

Outline • Seismic data • Seismic attributes • PCA • Image grand tour • Non-linear methods • Conclusions • Acknowledgements

Conclusions • The human is still the best interpreter we have • Attribute overload can overwhelm interpeters • Dimensionality reduction produces highly interpretable images

Acknowledgments • Prof. Kurt Marfurt (University of Oklahoma) • Mr. Victor Aarre (Schlumberger Norway Technology Center) • Mr. Tao Zhao (OU) • Dr. Marcilio de Matos (Petrobras) • CGG Veritas, Chesapeake Energy, Anadarko Petroleum, and the Government of New Zealand

Acknowledgments

Questions? bwallet@ou.edu http://geology.ou.edu/aaspi