During recent events, machine-learning focused Hackathon and Workshop, on July 19 and 26 respectively, it becomes apparent a common framework making ML relevant to geoscience will be very useful.

Here a first-iteration tabulated template* is intended for further discussion.

1) Data Prep and Processing -

AVO - say a simple two-term reflection amplitudes, R(𝜭) = P + G * sin2(𝜭), AVZ(𝜙) = AVO vs azimuth, where 𝜭 denotes angle of reflection (in subsurface, associated with offset in non-linear way, i.e., square of the sine of reflection angle), P zero-offset amplitude, G the gradient or slope, and 𝜙 azimuth (orientation at the surface)

PSDM - Kirchhoff / Beam / Wave Equation / Reverse-Time / Full Waveform Inversion

ML - Normalization / Standardization

2) Features Engineering -

Attributes - Amp, P, G, coherence, dip, density, velocity, porosity, Young's moduli, brittleness, etc.

PSDM - Anisotropy / attenuation Q, Velocity / moveout, Salt geometry as defined by top / base

ML - Dimensions reduction, Cluster analysis, Principal components, Eigenvectors

3) Model Fit

AVO / Earth model - Least squares, Sparse-spike, Max-likelihood

PSDM - Velocity-depth / Earth model / Inversion

ML - Logistic Regression / Support Vector Machine / Deep Neural Network / Convolutional Neural Network / Recurrent Neural Network

4) Predict

AVO / Earth model - Regression / Classification Type I to IV / Rock properties

PSDM - Reservoir shape and volume / Depth image / Illumination quality

ML - Regression / Classification / Probabilities (Facies, Flow Rate, Discrete Fracture Network)

Call for Action - share experience from recent projects, as well as perspective, after the virtual course "Applications of AI and machine learning for seismic reservoir characterization".


*abbreviations:

AVO - amplitude versus offset
ML - machine learning
PSDM - prestack depth migration