the next advancement in seismic interpretation, seg 2015
TRANSCRIPT
A Presentation by
Geophysical Insights, “From Insight To Foresight”, and Paradise are registered trademarks of Geophysical Research, LLC. All rights reserved. 2015.
The Next Advancement in Seismic Interpretation:
2015 SEG Annual Meeting, October 19-21
Rocky Roden
Sr. Consulting Geoscientist
Multi-Attribute Analysis lives hereTM
What is a Seismic Attribute?
• A measurable property of seismic data, such as
amplitude, dip, frequency, phase, and polarity.
Attributes can be measured at one instance of
time/depth or over a time/depth window, and may
be measured on a single trace, on a set of traces or
on a surface interpreted from seismic data. Schlumberger Oilfield Dictionary
• Seismic attributes reveal features, relationships,
and patterns in the seismic data that otherwise
might not be noticed. Chopra and Marfurt, 2007
Objective of using Seismic Attributes
• To take advantage of seismic attribute analysis and
today’s visualization technology and to mine
pertinent geologic information from an enormous
amount of seismic data.
• The ultimate goal is to enable the geoscientist to
produce a more accurate interpretation and reduce
exploration and development risk.
Most Common Seismic Attributes for Interpretation• Instantaneous Attributes
• Amplitude Defining Attributes
• Coherency/Similarity
• Curvature
• AVO Attributes
• Inversion
• Spectral Decomposition
Instantaneous AttributesReflection Strength (trace envelope, instantaneous amplitude)
Lithological contrasts
Bedding continuityBed spacingGross porosityDHIs
Instantaneous PhaseBedding continuity
Visualization of unconformities and faults
Instantaneous FrequencyBed thickness
Lithological contrasts
Fluid content (frequency attenuation)
Instantaneous Phase
Instantaneous Frequency
Seismicmicro.com
Seismicmicro.com
Amplitude Accentuating AttributesThese attributes help define how the amplitude stands out against surrounding
reflectors and background events.
Average EnergySweetness (frequency weighted envelope)RMSRelative Acoustic Impedance
DHI characteristicsStratigraphic variationsPorosityLithology variations
Average Energy Sweetness
Coherency/Similarity
Coherency, similarity, continuity, semblance and covariance are similar and relate to a measure of similarity between a number of adjacent seismic traces (multi-trace analysis). They convert data into a volume of discontinuity that reveals faults, fractures, and stratigraphic variations.
Crosscorelation-Based CoherenceSemblance-Based CoherenceVariance-Based CoherenceEigenstructue-Based CoherenceGradient Structure Tensor-Based CoherenceLeast-Squares-Based Coherence
Instantaneous Dip
Smoothed Similarity
Seismicmicro.com
Seismicmicro.com
AVO AttributesAVO attribute volumes are computed from prestack data (gathers) . They include combinations of near, mid, and far offset or angle stacks and depending on approximations of the Knott-Zoeppritz equations, various AVO components. Most of the AVO attributes are derived from intercept and gradient values or equivalents. Employed to interpret pore fluid and/or lithology.
Intercept (A)Gradient (B)Curvature (C)A*BA-C½ (A+B)½ (A-B)Far-Near(Far-Near)FarPoisson ReflectivityFluid Factor Lambda-Mu-Rho
Seismic Inversion Inversion transforms seismic reflection data into rock and fluid properties.
The objective of seismic inversion is to convert reflectivity data (interface properties) to layer properties.
To determine elastic parameters, the reflectivity from AVO effects must be inverted.
The most basic inversion calculates acoustic impedance (density X velocity) of layers from which predictions about lithology and porosity can be made.
The more advanced inversion methods attempt to discriminate specifically between lithology, porosity, and fluid effects.
Recursive Trace IntegrationColored InversionSparse SpikeModel-Based InversionPrestack Inversion (AVO Inversion)
Elastic ImpedanceExtended Elastic ImpedanceSimultaneous Inversion
Stochastic InversionGeostatisticalBayesian
Simultaneous P-Impedance Inversion
Russell and Hampson, 2006
Spectral DecompositionUse of small or short windows for transforming and displaying frequency spectra (Sheriff, 2005 Encyclopedic Dictionary of Applied Geophysics). In other words, the conversion of seismic data into discrete frequencies or frequency bands.
Layer thickness determinations
Stratigraphic variations
DHI characteristics (e.g. shadow zones)
Discrete Fourier TransformFast Fourier TransformShort Time Fourier TransformMaximum Entropy MethodContinuous Wavelet Transform
GaborGabor-MorleyGaussian
SpiceContinuous Wavelet Packet-Like TransformWigner-Ville DistributionSmoothed Wigner-Ville DistributionMatching PursuitExponential Pursuit
5 Hz
15.1 Hz11.4 Hz
8.7 Hz6.6 Hz
35 .5Hz
19.9 Hz
45.5 Hz 60 Hz
Original Full
Frequency
26.2 Hz
CurvatureCurvature is a measure of bends and breaks of seismic reflectors. Another way to describe curvature for any point on a seismic reflecting interface is the rate of change of direction of a curve.
FracturesFoldsFaults
Abele and Roden, 2012
Maximum Curvature
Mean CurvatureMaximum CurvatureMinimum CurvatureGaussian CurvatureMost Positive CurvatureMost Negative CurvatureShape Index CurvatureDip CurvatureStrike CurvatureCurvedness
New Terminology
– Entering the era of data deluge, where the amount of data outgrows the capabilities of query processing technology.
SEISMIC INTERPRETATION
TODAY
• Gathers• Large 3Ds with numerous processing versions• Numerous wells with associated data• Dozens to hundreds of seismic attributes
= terabytes of data
HOW DO GEOSCIENTISTS EFFICIENTLY INTERPRET ALL THIS
DATA AND UNDERSTAND HOW THESE DIFFERENT DATA
TYPES RELATE TO EACH OTHER?
– Uses computer algorithms that iteratively learn from the data and independently adapt to produce reliable, repeatable results.
Machine Learning employing current computing technology and visualization techniques addresses two significant issues in seismic interpretation:1.The Big Data problem of trying to interpret dozens if not
hundreds of volumes of data, and
2.The fact that humans cannot understand the relationship of several types of data all at once.
MACHINE LEARNING
Self-driving carsPractical speech recognitionEffective web searchImproved understanding of the human genome
Supervised Learning•Multi-layer Perceptron NN•Probabilistic NN•Generalized Regression•Radial Basis Function NN
Unsupervised Learning•K-means•Projection Pursuit•Principal Component Analysis•Self-Organizing Maps•Vector Quantization•Generative Topographic Mapping
Supervised Learning – the correct/desired answers are known from training on a dataset with known results, then this training is applied to a dataset with unknown results.
Unsupervised Learning – The correct/desired answers are not known from a previous dataset, the training adapts to the data, identifying natural patterns or clusters.
PRINCIPAL COMPONENT ANALYSIS (PCA)
A linear mathematical technique to reduce a large set of variables (seismic attributes) to a small set that still contains most of the variation in the large set.
PCA
Attribute A
Att
rib
ute
B
First Principal Component(largest variation)
Second Principal Component(2nd largest variation)
The eigenvector is the direction of the line showing the variance or spread in the data.
The eigenvalue is the value showing how much variance there is in the eigenvector.
WHAT IS THE LARGEST VARIATION IN THE DATA?
WHAT IS THE SECOND LARGEST VARIATION IN THE DATA?
How PCA relates to finding the most significant seismic attributes
(12 seismic attributes were employed)
FIRST PRINCIPAL COMPONENTTrace Envelope 25.17%SD Envelope 33.9 Hz 24.28%Sweetness 23.29%Average Energy 20.62%
SECOND PRINCIPAL COMPONENTDominant Freq 34.61%Instantaneous Freq 29.42%Instantaneous Q 17.68%
Highest Eigenvalue
Second Highest Eigenvalue
Highest Eigenvalues on each inline
The first principal component accounts for as much of the variability in the data as possible, and each succeeding component (orthogonal to each preceding) accounts for as much of the remaining variability as possible.
PRINCIPAL COMPONENT ANALYSIS (PCA)
The seismic attributes contributing the most to the first few principal components often indicate the most important seismic attributes to define geological features in the data.
BIOLOGICAL NEURAL NETWORK
The brain is a collection of 10 billion interconnected neurons, each is a cell that uses biochemical reactions to receive, process, and transmit information.
ARTIFICIAL NEURAL NETWORKS (ANN)
𝑎𝑖,1
Input Layer 1 Hidden Layer 2
𝑊𝑖,1 𝑎𝑖,2 𝑊𝑖,2 𝑎𝑖,3
An Artificial Neural Network is a computational simulation of a Biological Neural Network, composed of a large number of highly interconnected processing elements (neurons)
(An information processing technology pertaining to the area of machine learning in artificial intelligence)
SELF-ORGANIZING MAPS (SOM)
Self-Organizing Maps are artificial neural networks employing unsupervised learning methods.
A SOM is a cluster analysis and pattern recognition approach.
A SOM analysis produces a collection of neurons which classify data samples into categories, patterns or clusters based on their properties.
A neuron is a point that identifies a natural cluster of attributes.
Amplitude Spec. Decomp. Similarity Dip Azimuth Sweetness
SOM Classification for 10 Attributes
Each sample in the 3D survey will have 10 values associated with each attribute
Multi-Dimensional Space = Data from Numerous Attributeswe cannot visualize multiple dimensions at the same time
(10 attributes = 10 dimensional space)
SOM analysis maps the natural clusters from the numerous attributes’ data points to a 2D Colormap
Each neuron represents a cluster in the
data
3D SURVEY SPACE
ATTRIBUTE SPACE
SELF-ORGANIZING MAPS (SOM)
The neurons from the Classification on the 2D Colormapare interpreted for geologic significance
2D Colormap with 64 NeuronsNeurons Identifying Geologic Features
Seismic Attributes• Instantaneous• Geometric• AVO• Spectral
Decomposition• etc.
UnsupervisedLearning
PCASOM
Predictions• Facies• Porosity• Thickness• Faults/Fractures• DHI’s • Stratigraphy• Channels• Data Quality Artifacts• Pressure• etc.
Additional Information• Well Logs• Production• Interpretation Knowledge• etc.
CLASSIFIER
WELL WHAT’S YOUR PROBLEM ?
Fractures
Faults
Facies
WHICH ATTRIBUTES TO CHOOSE?
Select seismic attributes for SOM based on:
1. Principal Component Analysis
2. Previous knowledge and experience of appropriate attributes for the geologic feature of interest
Run SOM with combinations of attributes to resolve the defined geologic problem.
FOR EXAMPLE:Faults/Fractures Geometric Attributes (coherencies and curvatures)
Pore Fluids AVO, Inversion Attributes, etc.
A generic 8X8 (64 neurons) neuron count may be a good first run, however; depending on the scale of the geologic feature to identify, a lower count
such as 3X3 or 4X4 may be warranted.
REMEMBER, the higher the neuron count the more detail in the analysis
Interpret SOM results with 2D colormapIsolate neurons or combinations of neurons to
identify geologic featuresNE-SW trends
E-W trends
Shooting Direction
Classification at Top of Eagle Ford Shale(Geometric Attributes employed)
Nine geometric attributes employed in SOM analysis
Refine Interpretation:
• Modify attribute list employed in SOM analysis
• Use different neuron counts (3X3, 8X8, 20X20, etc.)
• Re-evaluate most meaningful neurons associated with geologic significant features (develop new 2D colormaps)
• Upthrown Fault Closure
• Approximately 100’ Reservoir Sand
• Two Producing Wells• #A-1 (gas on oil)
• #A-2 (oil)
3900’ Reservoir
Offshore Gulf of Mexico Case Study – Class 3 AVO
Previously 7 wells drilled in area, all wet or low saturation gas
GOAL: What DHI characteristics can be identified to refine reservoir
interpretation and employ in the region for further exploration.
Gulf of Mexico Case Study Seismic Attributes Employed in PCA
Acceleration of Phase Instantaneous Frequency Envelope Weighted
Average Energy Instantaneous Phase
Bandwidth Instantaneous Q
Dominant Frequency Normalized Amplitude
Envelope Modulated Phase Real Part
Envelope Second Derivative Relative Acoustic Impedance
Envelope Time Derivative Sweetness
Imaginary Part Thin Bed Indicator
Instantaneous Frequency Trace Envelope
First Principal Component
Eigenspectrum
Eige
nva
lues
Second Principal Component
Eigenspectrum
Eige
nva
lues
SOM ATrace EnvelopeEnvelope Modulated PhaseEnvelope 2nd DerivativeSweetnessAverage Energy
SOM BInstantaneous FrequencyThin Bed IndicatorAcceleration of PhaseDominant Frequency
Principal Component Analysis Inline EigenvaluesPrincipal Component Analysis
on 18 Instantaneous Attributes
SOM A Classification ResultsField
undrilled anomaly
N S
Attenuation
Original Stack Inline (dip) through field
SOM results – 5X5 colormap
SOM results – 4 neurons isolate attenuation
Attributes for Attenuation
1. Envelope Second Derivative
2. Envelope Modulated Phase
3. Trace Envelope
4. Average Energy
5. Sweetness
SOM A ClassificationField
undrilled anomaly
N S
Attenuation
Attributes for Attenuation
1. Envelope Second Derivative
2. Envelope Modulated Phase
3. Trace Envelope
4. Average Energy
5. Sweetness
SOM B Classification ResultsField
undrilled anomaly
N S
Oil/Water contact
Gas/Oil contact
Undrilled Flat Spot
Original Stack Inline (dip) through field
SOM results – 5X5 colormap
SOM results – 2 neurons isolate attenuation
Attributes for Flat Spots
1. Instantaneous Frequency
2. Thin Bed Indicator
3. Acceleration of Phase
4. Dominant Frequency
Field
undrilled anomaly
N S
Oil/Water contact
Gas/Oil contact
Undrilled Flat Spot
SOM B Classification
Attributes for Flat Spots
1. Instantaneous Frequency
2. Thin Bed Indicator
3. Acceleration of Phase
4. Dominant Frequency
• Today’s enormous amount of interpretation data and the desire to derive more meaningful information, requires implementation of New Technology.
• Machine Learning in the form of Principal Component Analysis (PCA) and Self-Organizing Maps (SOM) are resolving Big Data issues and extracting more meaningful information from our data.
• PCA identifies the most prominent attributes in our seismic data and helps determine appropriate attributes for SOM.
• SOM (ANN) identifies the natural patterns/clusters from various combinations of seismic attributes.
• These natural patterns/clusters help define geologic features that are difficult to interpret or can not be identified otherwise.
• Using modern computing and visualization technology, these Machine Learning approaches appear to be the Next Advancement in Seismic Interpretation.
Summary and Conclusions
THANK YOU!