[Paper Review] Propagating the prior from far to near offset: A self-supervised diffusion framework for progressively recovering near-offsets of towed-streamer data

Shijun Cheng, Tariq Alkhalifah|arXiv (Cornell University)|Feb 2, 2026

Seismic Imaging and Inversion Techniques0 citations

TL;DR

Authors present a self-supervised diffusion-based framework to reconstruct missing near-offset traces in towed-streamer marine seismic data by recursively extrapolating from far-offset data, with uncertainty estimates via ensemble sampling.

ABSTRACT

In marine towed-streamer seismic acquisition, the nearest hydrophone is often two hundred meter away from the source resulting in missing near-offset traces, which degrades critical processing workflows such as surface-related multiple elimination, velocity analysis, and full-waveform inversion. Existing reconstruction methods, like transform-domain interpolation, often produce kinematic inconsistencies and amplitude distortions, while supervised deep learning approaches require complete ground-truth near-offset data that are unavailable in realistic acquisition scenarios. To address these limitations, we propose a self-supervised diffusion-based framework that reconstructs missing near-offset traces without requiring near-offset reference data. Our method leverages overlapping patch extraction with single-trace shifts from the available far-offset section to train a conditional diffusion model, which learns offset-dependent statistical patterns governing event curvature, amplitude variation, and wavelet characteristics. At inference, we perform trace-by-trace recursive extrapolation from the nearest recorded offset toward zero offset, progressively propagating learned prior information from far to near offsets. The generative formulation further provides uncertainty estimates via ensemble sampling, quantifying prediction confidence where validation data are absent. Controlled validation experiments on synthetic and field datasets show substantial performance gains over conventional parabolic Radon transform baselines. Operational deployment on actual near-offset gaps demonstrates practical viability where ground-truth validation is impossible. Notably, the reconstructed waveforms preserve realistic amplitude-versus-offset trends despite training exclusively on far-offset observations, and uncertainty maps accurately identify challenging extrapolation regions.

Motivation & Objective

• Motivate the importance of near-offset data for processing steps like SRME, velocity analysis, and FWI.
• Develop a self-supervised framework that does not require near-offset ground-truth data for training.
• Leverage far-offset redundancy to learn offset-dependent seismic event patterns.
• Enable trace-by-trace recursive extrapolation from far to near offsets while preserving kinematics and amplitude.

Proposed method

• Use a conditional diffusion model that predicts the clean target patch x0 from a noisy input xt given a conditioning patch y.
• Train with overlapping patches that are shifted by a single trace to learn p(x0|y) without near-offset ground truth.
• Adopt x0-prediction in DDPMs for improved reconstruction quality.
• Perform trace-by-trace recursive inference from the nearest recorded offset toward zero offset using DDIM sampling.
• Provide uncertainty estimates via ensemble sampling from multiple diffusion realizations.

Experimental results

Research questions

• RQ1 Can a self-supervised diffusion model learn offset-dependent seismic patterns using only far-offset data?
• RQ2 Does recursive near-offset extrapolation from far to near offsets preserve kinematics and amplitudes compared to ground truth?
• RQ3 Can uncertainty maps from diffusion ensembles identify challenging extrapolation regions in near-offset reconstruction?
• RQ4 How does the proposed method compare to transform-based baselines like parabolic Radon transform in synthetic and field data?

Key findings

• The method achieves superior reconstruction fidelity over a parabolic Radon transform baseline on synthetic data.
• F-K spectra from reconstructions closely match ground-truth spectra, with reduced artifacts compared to baselines.
• Waveforms of near-offset traces are preserved with good amplitude and phase accuracy across traces from far to near offset.
• Uncertainty maps effectively identify regions with higher extrapolation difficulty, aiding downstream processing decisions.
• Field data results validate practical viability for real-world near-offset gaps where ground truth is unavailable.