Chaohua Liang | Searcharxiv

arXiv2026

Physics-Informed Neural Networks for Viscoacoustic Wave Propagation: Forward Modelling, Inversion and Discretization Sensitivity

Seismic wave forward and inverse modeling are fundamental tools for subsurface imaging and geological hazard assessment. Conventional grid-based numerical methods, such as finite-difference and finite-element approaches, often require dense discretization and repeated forward simulations, leading to high computational cost in inverse problems. Although deep learning has shown promise in seismic applications, its performance is commonly limited by the need for large labeled datasets and weak enforcement of physical constraints. In this study, we propose a unified physics-informed neural network (PINN) framework for forward modeling and parameter inversion of viscoacoustic wave propagation. By embedding the viscoacoustic wave equation into the learning process, the proposed framework accurately reproduces wavefields, attenuation, and phase characteristics, while enabling the simultaneous inversion of velocity and attenuation parameters from temporally sparse observations. Numerical experiments demonstrate that the PINN approach achieves stable and reliable accuracy compared with finite-difference solutions, while exhibiting reduced sensitivity to spatial discretization. These results highlight the potential of PINNs as a data-efficient and physically consistent alternative for high-resolution seismic modeling and inversion in attenuative media.

physics.geo-ph↗

arXiv2026

SeisBind: Physics-Aware Tri-Modal Representation Binding for Seismic Data via Contrastive Learning

This letter proposes a physics-aware multi-modal contrastive learning framework designed to transform complex seismic wavefields into human-readable physical representations. Traditional data-driven inversion methods often focus on pixel-wise mapping, which lacks physical grounding and interpretability. To address this, we introduce a novel framework that jointly aligns seismic shot gathers, subsurface velocity models, and explicit physical descriptors (e.g., mean velocity and gradients) in a shared latent space. By introducing these descriptors as a third modality, our approach encourages the learned embeddings to capture intrinsic geological semantics rather than superficial signal correlations. Experiments on the OpenFWI dataset demonstrate that the proposed method not only achieves robust seismic-to-velocity retrieval but also preserves meaningful physical semantics, enabling cross-modal inference of interpretable attributes. This representation-centric perspective provides a flexible foundation for expert-guided subsurface characterization.

physics.geo-ph↗