A novel prestack sparse azimuthal AVO inversion

TL;DR

This paper introduces a new sparse prestack azimuthal AVO inversion algorithm that combines a Euclidean prior model with machine learning techniques to improve seismic data interpretation, demonstrated on the Marcellus shale.

Contribution

It develops a novel Euclidean prior model for sparse and smooth reflectivity inversion and integrates machine learning with physical principles for azimuthal AVO analysis.

Findings

Effective clustering of attributes in the Marcellus shale.

Supports physical interpretability through Ruger model.

Enhances seismic interpretation with sparse, machine learning-based inversion.

Abstract

In this paper we demonstrate a new algorithm for sparse prestack azimuthal AVO inversion. A novel Euclidean prior model is developed to at once respect sparseness in the layered earth and smoothness in the model of reflectivity. Recognizing that methods of artificial intelligence and Bayesian computation are finding an every increasing role in augmenting the process of interpretation and analysis of geophysical data, we derive a generalized matrix-variate model of reflectivity in terms of orthogonal basis functions, subject to sparse constraints. This supports a direct application of machine learning methods, in a way that can be mapped back onto the physical principles known to govern reflection seismology. As a demonstration we present an application of these methods to the Marcellus shale. Attributes extracted using the azimuthal inversion are clustered using an unsupervised learning algorithm. Interpretation of the clusters is performed in the context of the Ruger model of azimuthal AVO.