This is the first study to formulate the design of a seismic reconstruction operator as a constrained, LLM-driven program evolution task and proposes SeisEvo (Seismic Algorithm Evolution), which does not optimize a single reconstruction result but searches for the algorithm that produces it.
Abstract
Classical seismic data reconstruction relies on manually designed structural priors and iterative operators, whose coupled design space is far larger than manual trial and error can explore systematically. Deep-learning methods encode the reconstruction rules in learned weights rather than in an explicit operator that can be inspected and modified. We propose SeisEvo (Seismic Algorithm Evolution), which does not optimize a single reconstruction result but searches for the algorithm that produces it. Starting from a classical reconstruction algorithm, an LLM-driven multi-agent search modifies only the components that the user has opened for editing, without prescribing the mechanism to be discovered. Candidates that violate the physical constraints of the task are rejected outright, and the remaining ones are scored by execution. The output is neither an agent system nor a neural network, but a standalone white-box algorithm that requires no agent or neural network at inference time. For interpolation without added noise, the search discovered a residual-gated, phase-aligned dip-consistency projection; Evo-POCS improves the SNR over classic POCS by 3.49 dB on average across missing ratios from 30% to 70%. For simultaneous interpolation and denoising, it discovered a reliability-grouped singular-value shrinkage; Evo-MSSA improves the average reconstruction SNR by more than 7 dB over classic MSSA and by more than 3 dB over a stronger rank-reduction baseline. Both operators retain their gains on data not used during the search. To the best of our knowledge, this is the first study to formulate the design of a seismic reconstruction operator as a constrained, LLM-driven program evolution task. Agentic algorithm evolution can thus complement deep learning in discovering explicit, inspectable, and deployable seismic processing algorithms.
Neural networks for seismic fault segmentation are often borrowed from computer vision and medical imaging domains where they train under relatively much larger labeled data resources. Optimizing their architecture under tight labeled data budgets as are common in geophysical applications is not a trivial problem. Manually designing data-optimal architectures is time-consuming while classical neural architecture search (NAS) is restricted to hand-crafted search spaces and large compute budgets. We present an agentic NAS system in which a panel of three large language models (Claude, GPT-5.1, and Gemini~2.5~Pro) debates each candidate architecture to unanimous consensus, authors the complete PyTorch implementation, cross-reviews it, and submits it to an automated validate-train-score loop with a hard 450K parameter budget, keep-or-revert lineage, and a memory of failed mechanisms. Operating on source code rather than a predefined operation menu, the search ran on a single consumer GPU and trained only eight candidates. It discovered \ours{}: a 425K-parameter encoder-decoder with a strip-pooling bottleneck, squeeze-and-excitation gating, an asymmetric one-conv decoder, and a feature-pyramid fusion neck. Trained under a protocol identical to all baselines on sections derived from the Thebe fault dataset, it attains the highest F1 (0.578) and IoU of all models tested while being the smallest, outperforming a published-capacity U-Net (31M parameters, F1 0.484), DeepLabV3-ResNet50 (39.6M, 0.516), an Attention U-Net(1.83M, 0.502). The search cost 101 LLM calls ($\sim$1.15M input / 0.39M output tokens) and roughly one GPU-day, making consensus-gated LLM panels a practical, low-cost route to domain-specific architecture discovery.
We propose a variational autoencoder framework to directly assess uncertainties in subsurface models produced by single‐ and multiparameter full waveform inversion (FWI). The new method does not require pretraining on labeled data, thus it significantly reduces computational cost and storage requirements. The framework takes the seismic shot gathers as input and returns a set of possible velocity models that fit the input data. The network has three key components: (a) an encoder that maps the input seismic shot gathers to feature distributions in the latent space, (b) a latent vector sampled directly from feature distributions, and (c) a decoder that maps the sampled latent vector to the velocity model space. We incorporated Deep Image Prior principles, leveraging convolutional layers and LeakyReLU activations to regularize the inversion and improve reconstruction. At each epoch, the reconstructed velocity model is passed to a finite difference partial differential equation solver for forward modeling. We then calculate the data misfit between the input seismic data and the simulated data. Additionally, we compute the FWI gradient by cross‐correlating the adjoint and forward wavefield; the FWI gradient is then back‐propagated to guide the update of the neural networks weights and biases. The injection of physics‐based gradients further constrains the inversion and enhancing convergence. We tested the framework on synthetic acoustic and elastic seismic data sets, inverting models of varying complexity and quantifying their associated uncertainties. The results show accurate mean velocity models with meaningful uncertainty estimates, highlighting the potential of the proposed method for practical FWI applications.
A. Elmeliegy, Mrinal K. Sen, A. Dhara et al.· Journal of Geophysical Resea...· 0 citations
Coseismic permanent ground displacement is essential for understanding earthquake rupture mechanisms and assessing seismic hazards. However, traditional geodetic methods are often limited by sparse spatiotemporal coverage and high costs. Strong-motion records offer a valuable supplement but are frequently contaminated by baseline offsets. Existing baseline correction methods largely rely on visual inspection and empirical image-based judgments, lacking a clear physical foundation and sufficient robustness. As a result, they are ill-suited for the automated processing of large-scale strong-motion datasets, and the valuable displacement information contained in these records remains underutilized. This study breaks away from conventional approaches by reformulating baseline correction as an inverse problem in robot trajectory planning and proposing a fully automated correction framework based on jerk constraints. The method achieves fully automated processing and effectively handles challenging scenarios such as velocity pulses, complex waveform distortions, and nonideal instrument states. Its strong cross-event adaptability is validated using records from the Chi-Chi, Wenchuan, and Kahramanmaraş earthquakes. By providing a physical interpretation of baseline offsets, this work eliminates the dependence on empirical, image-based adjustments and offers a novel, physically grounded approach for strong-motion signal processing.
Xiao-yu Chen, Ming-Yang Jin, Zhibin Sun et al.· Bulletin of The Seismologica...· 0 citations
Seismic wavefield simulation is fundamental to seismology, but conventional finite-difference (FD) methods remain limited by numerical dispersion and stability constraints, which often require dense spatial grids and small time steps and thereby severely limit the effectiveness of iterative inversion workflows. We introduce a conditional diffusion-based wavefield propagator that advances seismic wavefields recursively from one time step to the next. Instead of learning an unconditional data distribution of wavefield evolution, the model is conditioned by a short history of recent wavefield time steps (snapshots), the velocity model, and the wavefield time step index, allowing it to represent the conditional transition between adjacent physical states. By training the network to directly predict the clean next wavefield snapshot, this strong physical conditioning makes it possible to replace the iterative reverse diffusion process with a single network evaluation for each predicted snapshot. To improve stability over long recursive rollouts, we further introduce a causal time-weighted loss, in which adaptive weights, accumulated as exponential moving averages of per-snapshot training errors, emphasize training directions that are consistent with the forward propagation sequence and reduce the amplification of one-step prediction errors. Because the learned propagator is tied to the temporal spacing of the training snapshots rather than to the FD stability limit, it can advance the wavefield using a physical time step ten times larger than that required by the underlying solver. Experiments on the Overthrust, SEG/EAGE, and Marmousi models show that the proposed method accurately reproduces wavefield snapshots and shot gathers and achieves an end-to-end speedup of 2.17 x over a GPU-accelerated tenth-order staggered-grid FD implementation under matched hardware conditions.
Tying seismic data to well logs is a critical step in seismic amplitude processing and interpretation. This process is often repeated throughout the exploration phase to enhance the understanding of well characteristics, such as the time-depth relationship. However, the presence of data noise and inherent uncertainties makes the well-tie task time-consuming and complex. It requires the estimation and careful control of several parameters to ensure reliable results. We propose a workflow based on segmented global optimization that partially automates the well-seismic tying process by estimating the matching position between the synthetic and seismic traces and automatically aligning them. The matching position is estimated using Time-Weighted Dynamic Time Warping (TWDTW), which accounts for the maximum allowable perturbation in the velocity log. In conjunction with TWDTW, a global Bayesian optimizer estimates the tie-related parameters. Following this step, the alignment is performed through an optimization-based method that incorporates segmentation to estimate both the velocity log perturbation and wavelet phase, constrained by a predefined tolerance. The well logs and seismic traces are segmented, and each segment is aligned using constrained Dynamic Time Warping (DTW). This segmentation increases the number of knots available for interpolating the velocity perturbation curve, resulting in a more detailed and refined alignment. The proposed partial automation improved the correlation in all six wells evaluated, with four wells achieving correlation values above 70%, without introducing physically unrealistic distortions.
Rafael da Costa Silva, Marcelus Glaucus de Souza Araújo, Luiz Antonio Rozendo et al.· Geophysics· 0 citations
Full Waveform Inversion (FWI) is a powerful tool for subsurface velocity reconstruction but remains highly ill-posed, sensitive to acquisition limitations, often requiring some form of regularization to reduce artifacts and enhance resolution. While recent developments have shown that generative models can inject learned priors directly into the FWI optimization process, such approaches typically require additional, computationally expensive inversion iterations. In this study, we propose a post-FWI refinement strategy based on a Flow Matching (FM) generative model, which leverages learned geological priors without re-running FWI. The method guides the deterministic generative process using the FWI result, as well as well logs, if available. Synthetic and field data experiments demonstrate that we can inject well information and our geological expectations (prior) into the provided FWI result, and thus, we can effectively enhance its resolution and geological quality. In fact, the well prior even managed to alter the model depth to fit the well information, which is a form of correcting for depth misties.
