SPE-154215 Combining Absorption and AVO Seismic Attributes Using Neural Networks to High-Grade Gas Prospects A. Rahimi Zeynal, University of Southern California F. Aminzadeh, University of Southern California A. Clifford, Saratoga Resources, Inc.
Statement of Problem: identifying possible commercial shallow gas targets for drilling
Statement of Problem: Naturally-occurring gas seeps have long been known to exist at Grand Bay field located in South- Eastern Louisiana.
Need for improvement: - Well logs: small investigation radius, high resolution - Seismic data: large investigation radius, low resolution Solution Methods: - 1989, (S.R. Rutherford), AVO variations in gas sands - 2004, (J. Walls) Reducing hydrocarbon indicator risk - 2011, (A. Clifford) Detecting gas using AVO attribute analysis
Suggested approach: AVO Attributes Absorption Attributes Gas Volume Well Logs
workflow for gas detection: First Step Create AVO attribute (Amplitude Variations with Offset)
AVO Attribute Analysis: AVO is a hydrocarbon indicator that is widely accepted as a means of detecting gas saturated sandstones. Sharp drop in Vp with small increase of gas saturation
AVO Attribute Analysis: - Brine saturated case will show a decaying amplitude with offset Offset - Gas saturated case will show an increasing amplitude with offset
AVO Attribute Analysis:
AVO Attribute Analysis: Near Offsets Mid Offsets Far Offsets
900 Sand Pre-Stack Time Migrated (PSTM) far minus nears AVO anomaly 1050 Sand Pre-Stack Time Migrated (PSTM) far minus nears AVO anomaly
workflow for gas detection: Second Step Calculate frequency dependent attributes
Frequency Attribute Analysis: High frequency content of seismic response attenuates more extensively as it propagates through gas-bearing reservoirs. ATTRIBUTE Average Frequency Squared (AFS) Frequency Slope Fall (FSF) Spectrum Area Beyond MDA (SAB) Dominant Frequency (MDA) Avbsorption Quality Factor (AQF) FUNCTION Magnifies high frequency loss Highlights flattening of spectrum A measure of high frequency loss Reduces the impact of noise Overall measure of absorption
Frequency Attribute Analysis: Dominant Frequency MSA*Dominant Frequency Absorption Quality Factor Average Frequency Squared Frequency Slope Fall
Frequency Attribute Analysis: AQF is defined as the area of the power spectrum beyond the dominant frequency.
AQF attribute anomalies for 900 Sand, Grand Bay Field. AQF attribute anomalies for 1050 Sand, Grand Bay Field
AVO AQF AVO AQF
workflow for gas detection: Third Step Create required logs from well data
Well logs and gas effects: Gas has a very marked effect on both density and neutron logs. It will result in a lower bulk density, and a lower apparent neutron porosity
workflow for gas detection: Fourth Step Train a neural network based on the attributes and well logs
Neural Network Training:
Neural Network Training: The Neural Network were trained based on seed points from well control data, AVO and absorption related attributes. Training Inputs - Gas and background pick sets from AQF and AVO gas bearing points - Using neutron, density and gamma ray logs Output - Gas Cube
workflow for gas detection: Fifth Step Create the gas probability volume
Neural Network property prediction: The real power of the reliability of results comes from combining absorption and AVO anomalies plus log data in evaluating the impact of different concepts on the output. Neural network-based methods Solve the non-linear relationship between the seismic data and reservoir properties
ANN log training based on multi-log combinations plus AVO/frequency attributes for 900 sand. ANN log training based on multi-log combinations plus AVO/frequency ANN log training based on multi-log combinations plus AVO/frequency attributes for 1050 sand
Conclusion: 1- Artificial Neural Network is a successful tool to detect shallow gas sands with 3D seismic data. 2- Training the Neural Network with AVO & Absorption Quality Factor (AQF) attributes results in better gas sand identification. 3- Neural Network yielded highest resolution when using AVO and AQF attribute in conjunction with Well logs. 4- The analysis supports the presence of numerous undeveloped pockets of shallow gas in the field as well as identifying new possible leads.
The authors would like to express their gratitude for use of dgb sopendtect and SMT s Kingdom Suite software to generate the results in this paper.
Thanks for Your attention