Machine learning-based comparative assessment of seismic inversion approaches for porosity estimation

Abstract

Seismic inversion is a fundamental technique for extracting quantitative subsurface properties from seismic data. This study confidently presents a thorough comparative assessment of five seismic inversion methods: Coloured Inversion (CI), Band-Limited Inversion (BLI), Model-Based Inversion (MBI), Maximum Likelihood Inversion (MLI), and Sparse Layer Reflectivity (SLR), all enhanced by a Probabilistic Neural Network (PNN) for porosity estimation. Using seismic data from the Blackfoot field in Canada, we rigorously evaluate the effectiveness of these techniques in delineating reservoir zones and accurately predicting porosity. Our compelling results demonstrate that MBI, BLI, and CI significantly outperform the other methods in identifying reservoir boundaries and aligning closely with well-log data. When integrated with PNN, MBI and BLI excel in mapping reservoir porosity, providing reliable predictions ranging from 5 to 20% and impedance values between 7000 and 18,000 m/s*g/cc. These integrated approaches deliver an unparalleled understanding of reservoir characteristics, far surpassing the precision and reliability of traditional methods. This research underscores the immense potential of combining advanced seismic inversion techniques with neural network models to revolutionize reservoir characterization. Our findings firmly establish the effectiveness of MBI and BLI, particularly when paired with PNN, in delivering accurate and high-resolution insights into subsurface porosity and reservoir boundaries. © Indian Academy of Sciences 2025.