arXiv subjects

Bao Deng

Publications and source records attributed to Bao Deng.

2 recordsLinked to original sources

arXiv2026

Accurate Interpolation of Ambient Noise Empirical Green's Functions by Denoising Diffusion Probabilistic Model and Implicit Neural Representation

Empirical Green's functions (EGFs) extracted from seismic ambient noise have been widely used to image Earth's interior structures, and the resolution of EGF-based tomography depends on the spatial density of seismic stations. However, due to cost and logistical constraints, it is often difficult to deploy dense seismic networks suitable for high-resolution tomography. While reliable interpolation of EGFs at unsampled locations could enhance tomographic resolution, the task remains inherently challenging and underexplored due to the dispersive nature of EGFs. In this study, we introduce DIER (diffusion-assisted implicit EGF representation), a self-supervised learning framework that integrates implicit neural representation with denoising diffusion probabilistic models to achieve high-fidelity EGF interpolation. In DIER, the diffusion process is conditioned on station coordinates to guide the transformation from random noise into EGF waveforms, which allows flexible reconstruction of five-dimensional EGF fields without labeled data or synthetic waveforms. We demonstrate the effectiveness of DIER through continent-scale EGF interpolation across the United States. The results show that DIER significantly outperforms the conventional radial basis function-based interpolation approach by generating EGFs with markedly improved phase alignment and dispersion characteristics. Surface wave tomography using the phase velocities derived from the interpolated EGFs also closely matches a reference model constructed from data acquired by a much denser seismic network. Our findings suggest that DIER provides a promising and cost-effective approach toward high-resolution ambient noise tomography in regions with sparse station coverage.

physics.geo-ph↗

arXiv2025

DispFormer: A Pretrained Transformer Incorporating Physical Constraints for Dispersion Curve Inversion

Surface wave dispersion curve inversion is crucial for estimating subsurface shear-wave velocity (vs), yet traditional methods often face challenges related to computational cost, non-uniqueness, and sensitivity to initial models. While deep learning approaches show promise, many require large labeled datasets and struggle with real-world datasets, which often exhibit varying period ranges, missing values, and low signal-to-noise ratios. To address these limitations, this study introduces DispFormer, a transformer-based neural network for $v_s$ profile inversion from Rayleigh-wave phase and group dispersion curves. DispFormer processes dispersion data independently at each period, allowing it to handle varying lengths without requiring network modifications or strict alignment between training and testing datasets. A depth-aware training strategy is also introduced, incorporating physical constraints derived from the depth sensitivity of dispersion data. DispFormer is pre-trained on a global synthetic dataset and evaluated on two regional synthetic datasets using zero-shot and few-shot strategies. Results show that even without labeled data, the zero-shot DispFormer generates inversion profiles that outperform the interpolated reference model used as the pretraining target, providing a deployable initial model generator to assist traditional workflows. When partial labeled data available, the few-shot trained DispFormer surpasses traditional global search methods. Real-world tests further confirm that DispFormer generalizes well to dispersion data with varying lengths and achieves lower data residuals than reference models. These findings underscore the potential of DispFormer as a foundation model for dispersion curve inversion and demonstrate the advantages of integrating physics-informed deep learning into geophysical applications.

physics.geo-ph↗