Model-based seismic inversion for unmapped subsurface faults using ambient noise tomography

Project Abstract

This study presents a model-based seismic inversion framework that leverages ambient noise tomography to map unmapped subsurface faults with high spatial fidelity. By exploiting the pervasive, low-amplitude seismic energy present in the near-surface environment, we develop a robust inversion scheme that extracts phase and amplitude information from cross-correlated ambient seismic records to reconstruct S-wave and P-wave velocity perturbations associated with fault zones. The methodology integrates an efficient Green’s function estimation approach with a hierarchical Bayesian inversion to jointly constrain fault geometry, fracture density, and velocity contrasts, while explicitly accounting for noise, anisotropy, and potential nonlinearity in the subsurface response. We begin with a data-driven ambient noise preprocessing pipeline that includes spectral whitening, temporal normalization, and cross-correlation stacking to generate empirical Green’s functions between strategically deployed sensor pairs. Next, we implement a multiscale inversion strategy that progressively recovers large-scale fault features before refining with high-resolution local perturbations, thereby mitigating nonuniqueness and promoting stable convergence. The forward model couples 3D acoustic-elastic wave propagation with anisotropic parameterization to simulate the ambient field response for a given fault model, enabling synthetic experiments that guide parameterization choices and prior formation. Regularization terms incorporate geological plausibility through fault plane orientation constraints, fracture density limits, and smoothness priors consistent with regional tectonics. To quantify uncertainty, we adopt a hierarchical Bayesian framework that yields posterior distributions for velocity perturbations, fault geometry, and inversion hyperparameters, providing credible intervals and model evidence for model comparison. The study encompasses comprehensive synthetic tests to evaluate sensitivity to data coverage, noise levels, and sampling geometry, followed by application to field datasets from tectonically active regions where unmapped faults are suspected but not yet characterized by conventional active-source surveys. We demonstrate that ambient noise tomography, when embedded in a model-based inversion with explicit misfit definitions for phase and amplitude spectra, can reveal subtle velocity contrasts and discontinuities indicative of fault zones, including hidden or dislocated segments. Results show improved localization of fault boundaries, enhanced resolution of fracture zones, and coherent integration with existing geological information, offering a noninvasive alternative for fault mapping in urban, offshore, and environmental monitoring contexts. The workflow is validated against borehole logs, microseismic catalogs, and independent geophysical imaging, demonstrating consistency across multiple scales and datasets. Practical considerations are discussed, including sensor deployment strategies, data duration requirements, computational cost, and scalability to large-area surveys. The outcomes advance the capability to delineate unmapped subsurface faults, support hazard assessment, and inform subsurface risk management by delivering probabilistic fault models that integrate seamlessly with geotechnical and seismic hazard analysis.

Project Overview

What This Project Is About

---

A plain-language overview of using ambient seismic noises to infer hidden faults underground by building and testing models that explain how sound waves travel through the Earth.

---

---

The Problem It Addresses

---

Often faults are not visible at the surface, making it hard to assess earthquake risks. Traditional methods can be expensive or limited by data availability. This project explores a low-cost way to detect and map unmapped faults using naturally occurring seismic noise and computer models.

---

---

Objectives of the Project

---

1. Learn the basics of seismic waves and ambient noise data.
2. Build a simple model that links wave behavior to underground faults.
3. Develop a workflow to invert observed data into a fault map.
4. Test the method on synthetic data and, if possible, real data.
5. Assess the method’s limitations and potential improvements.

---

---

What You Will Do Step by Step

---

1) Review basic seismology concepts and ambient noise. 2) Create or obtain a small dataset of background seismic signals. 3) Build a basic forward model that predicts signals for given subsurface faults. 4) Apply a simple inversion to estimate fault locations. 5) Validate results with known references or simulations. 6) Discuss uncertainties and practical uses.

---

---

Expected Outcome

A clear, tested approach that uses ambient noise to indicate where unmapped faults may lie, plus a discussion of when it works best and how accurate the results are. This could guide further fieldwork or more detailed studies.