Paper submitted to IMAGE 2026, by Graeme Stock, Olga Brusova, Mark Roberts, David Brookes, Amy Knowles, Sahil Mannick, Rebecca Yates (TGS).
Abstract
Reprocessing multi-vintage marine seismic data to a consistent broadband standard remains a major technical and operational challenge, particularly when acquisition parameters vary significantly between surveys and when reliable acquisition auxiliary data are incomplete or unavailable. One of the most critical steps in modern marine seismic reprocessing is deghosting, where surface-related ghost reflections generated at the sea surface are removed to recover low- and high-frequency bandwidth, stabilize amplitudes, and improve temporal resolution. This abstract presents an augmented machine-learning (ML) strategy for deghosting legacy marine seismic data, demonstrated on a large multi-survey reprocessing project from the Krishna– Godavari (KG) Basin offshore India. The proposed workflow combines the robustness and efficiency of a generalized ML deghosting model with survey-specific adaptation derived from limited, high-quality deterministic deghosting results. The augmented approach delivers stable broadband spectra, improved seismic image quality, and a high degree of consistency across surveys acquired with differing tow depths and acquisition geometries, while maintaining favorable turnaround times for large reprocessing projects.
Read the full article here.