Robust joint imaging and tomographic Q-estimation based on full wavefield matching using a machine learning constraint

Abstract

Seismic wave attenuation, quantified by the quality factor (Q), leads to energy loss and waveform distortion, significantly degrading seismic data quality and resolution. Accurate Q estimation is essential for understanding subsurface properties, particularly in applications such as carbon capture and storage (CCS) and near-surface studies, where attenuation effects are pronounced due to the presence of fluids, gases, or loose soil. Traditional Q tomography methods predominantly rely on spectral-ratio or centroid-frequency shift approaches to account for attenuation effects. However, these methods often face significant limitations, including oversimplified wave propagation assumptions, poor localization in heterogeneous media, and a tendency to produce smeared results, ultimately reducing resolution and accuracy.

To address these challenges, we introduce a novel Q-estimation approach that integrates full-waveform matching for accurate attenuation-effect estimation and compensation during the migration process. The Full Wavefield Migration method is enhanced by incorporating Q into a one-way modeling operator, utilizing full-waveform matching for precise Q estimation, and applying a Random Forest regression constraint to mitigate cross-talk between Q and reflectivity. This approach enables robust and localized Q estimations. Numerical examples demonstrate its effectiveness in accurately retrieving both reflectivity and attenuation models, thereby improving imaging resolution in complex subsurface environments.