Unsupervised erratic noise suppression in prestack seismic gathers via PnP-ADMM with implicit neural representation

Abstract Deepwater basins have become major targets for commercial oil and gas exploration. However, as exploration advances into ultra-deepwater basins, technical challenges become increasingly severe, as exemplified by the Liwan Sag in the northern South China Sea (SCS). The rugged seafloor terrain and great water depth degrade seismic imaging quality and increase uncertainties in drilling and geological interpretation. Conventional towed streamer reflection seismic data provide dense reflection information but are limited by relatively narrow observation angles and limited offset coverage. In contrast, ocean-bottom seismometer (OBS) data provide first-arrival traveltimes at larger offsets, thereby complementing the limited offset coverage of streamer data. In this study, we collate and preprocess joint OBS and towed streamer observations from the Liwan Sag to obtain seismic data with an improved signal-to-noise ratio for velocity model building over the full depth range. We establish an integrated workflow with three stages: seawater velocity model building, subseafloor background velocity model building, and subseafloor velocity refining. The resulting velocity models are evaluated using traveltime fitting, migration image quality, and common image gather (CIG) flattening, and are then used for imaging in the depth domain and geological interpretation of the survey line. This case study provides a practical workflow that integrates seismic data preprocessing, full-depth velocity model building, and stratigraphic imaging for ultra-deepwater seismic exploration.