Geophysics and Tectonics Seminar - winter-2025

Jan. 15, 2025
noon - 1 p.m.
Online: https://ucla.zoom.us/j/91518017395

Presented By:

• Prof. Sylvain Barbot - USC

The mechanics of slow-slip events and earthquakes is controlled by the constitutive behavior of rocks in active fault zones, which is sensitive to many factors encompassing lithology, temperature, confining and pore-fluid pressure, and slip-rate, among others. Understanding the frictional properties of faults is crucial to predicting many aspects of the seismic cycle, from the source characteristics and recurrence patterns of earthquakes to the mechanics of remote triggering. Here, we describe a constitutive model that explains the slip-rate-, state-, temperature-, and normal-stress-dependence of fault friction for a wide variety of rock types, explaining the evolution of frictional stability under various barometric and hydrothermal conditions relevant to natural and induced seismicity, encompassing the brittle-ductile transition. The frictional strength is controlled by the area of contact junctions that form along a rough interface or by grain-to-grain contact in fault gouge and follows a nonlinear function of normal stress. The physical model explains the direct and evolutionary effects following perturbations in temperature, normal stress, and slip-rate, and the dependence of the frictional parameters on ambient physical conditions. The competition among healing and deformation mechanisms explains the dependence of fault stability on temperature, slip-rate, and effective normal stress for a wide range of rocks. The brittle-to-flow transition at the bottom of the seismogenic zone is caused by the thermobaric activation of semi-brittle deformation mechanisms. The model unifies and extends previous formulations, providing a single framework to explain rock deformation in Earth's brittle and ductile layers. We illustrate an application to Southern California geology, integrating the frictional properties of schists, granodiorite, diorite, granite, gabbro, and olivine, representative of the continental crust and upper mantle, into numerical simulations of the seismic cycle.

Jan. 22, 2025
noon - 1 p.m.
Geology 1707

Presented By:

• Kyungjae Im - USC

Much progress has recently been made in the development of stress-based models for forecasting earthquakes. However, forecasting magnitudes in complex fault geometry remains a challenge. Quake-DFN, an open-source earthquake simulator, was developed to address this challenge. It allows simulating sequences of earthquakes in a 3-D Discrete Fault Network governed by rate and state friction, a phenomenological law established based on laboratory observations. I first show simulation results with a tectonically loaded complex fault system in this talk. In particular, we select Ridgecrest earthquake fault geometry and show how the initial stress field alters rupture sequences. We then investigate the maximum magnitude of induced earthquakes using planar fault geometry using varied initial stress and injection rates. We define the radii of two different slip modes, aseismic (Ra) and seismic slip (Rs), and derive an expression for maximum magnitude evolution.

Jan. 29, 2025
noon - 1 p.m.
Geology 1707

Presented By:

• Dr. Abdulfattah Aldajani - Massachusetts Institute of Technology

Here are essentially two principal uses of seismic data, namely, obtaining an image of the earth’s subsurface and estimating elastic properties of the subsurface rock layers among others. While imaging is focused at obtaining the spatial locations of interfaces, seismic inversion is aimed at converting the interface information between subsurface layers to critical interval information about the target formation. The goal of seismic reflection inversion is to estimate earth model parameters from remotely sensed geophysical data. Seismic reflection data, which has traditionally provided structural definition of reservoirs, is being exploited for more extensive and accurate reservoir characterization for exploration prospecting, and field development in both conventional and unconventional plays. This means that in addition to delineating the target reservoir structure and geometry, it is as critical to effectively utilizing the seismic reflection wave field to predict the elastic parameters and the mechanical properties of the target formation and infer critical information about the lithology such as the facies, anisotropy, natural fractures, and fluid content among others. In order to achieve such a goal, seismic inversion is an important processing step that uses the surface seismic reflection amplitudes, integrated with other data types, namely, well logs, well test, geology etc.

Feb. 5, 2025
noon - 1 p.m.
Geology 1707

Presented By:

• Peter Gerstoft - University of California San Diego

Physics-Informed Neural Networks (PINN) can be used for sound field reconstruction from a limited number of observations (interpolation). PINN can also be used for solving the inverse problem of extracting physical properties from observed vibrations. PINN has emerged as the interface between machine learning methods and physics-based models, supplementing the information content from limited data with constraints based on physical models to provide physically viable predictions. Hence, PINNs are well adapted for robust interpolation of the sound fields from sparse recordings without requiring training data.

Feb. 12, 2025
noon - 1 p.m.
Geology 1707

Presented By:

• Prof. Gareth Funning - UCR

Fault creep - slow sliding of faults in the absence of seismic shaking - is observed on several faults in California. It plays a role in reducing the strain that accumulates on those faults, and may also inhibit seismic ruptures. By mapping the extents of creep on faults we can better constrain the areas that are not creeping, and therefore likely to rupture in future, and also potentially identify the geological conditions responsible. In this talk, I will present recent InSAR, seismic and modelling results that suggest that creep may be even more extensive in northern California than previously thought.

Feb. 19, 2025
noon - 1 p.m.
Geology 1707

Presented By:

• Hojjat Kaveh - Caltech

Earthquake forecasting remains a formidable challenge due to the complex and chaotic nature of fault systems. Earthquake sequences exhibit spatiotemporal chaos, where small perturbations can lead to vastly different outcomes, complicating predictability. A particularly significant challenge is forecasting extreme events—large earthquakes that deviate from the system’s typical behavior and carry a disproportionately high impact. This seminar explores recent advances in tackling these challenges through dynamical systems theory, data assimilation, and reduced-order modeling. We discuss how chaotic attractors can help constrain feasible initial conditions for forecasting, how optimization methods can identify pre-event states, and how data assimilation techniques can incorporate low-resolution and noisy observations into models. We also examine the role of machine-learned reduced-order models in bridging the gap between theoretical understanding and practical forecasting applications. Through some case studies, we highlight promising pathways for improving earthquake forecasts while acknowledging the remaining hurdles in observational constraints and computational modeling.

Feb. 26, 2025
noon - 1 p.m.
Geology 1707

Presented By:

• Abhijit Ghosh - UCR

Artificial intelligence (AI) and machine learning (ML) are revolutionizing earthquake science by enabling efficient analysis of complex datasets. These techniques enhance our understanding of fault dynamics, seismic hazard assessments, and earthquake predictability. This talk explores two ML applications in earthquake physics: (1) AI-driven analysis of rupture propagation in the San Andreas Fault (SAF) system and (2) AI-enhanced earthquake detection for improved seismic catalogs. The first study investigates rupture pathways in the SAF system, a critical challenge due to complex fault geometry, stress conditions, and frictional properties. Traditional methods rely on computationally expensive iterative simulations with limited generalizability. To address this, we employ ML models trained on Rate-State earthquake simulator (RSQSim) outputs to assess key parameters influencing rupture behavior at the San Andreas-Garlock fault intersection. Our results highlight the dominant role of pre-earthquake fault conditions—especially those on the nucleating fault—in determining rupture paths. This ML-based approach offers a scalable tool for rupture propagation studies and seismic hazard assessments. The second study focuses on next-generation earthquake catalogs using AI-driven detection algorithms. We introduce AI-PAL, a deep learning workflow that integrates a Self-Attention RNN (SAR) model with PAL (Pick, Association, Location) detections, a rule-based method. When applied to seismic datasets from two tectonically active regions, AI-PAL enhances detection completeness, processing speed, and temporal stability. Compared to existing models, it identifies more seismic events with higher accuracy, offering insights into fault behavior such as distributed seismicity, persistent asperities, and rupture barriers. These case studies demonstrate AI/ML’s transformative impact on earthquake science. They enable scalable rupture analysis, improved seismic hazard models, and enhanced earthquake detection capabilities.

March 5, 2025
noon - 1 p.m.
Geology 1707

Presented By:

• Paul Richardson - CGS

Wildfires and consequent postfire geohazards, specifically runoff-induced debris flows, are a major threat to California communities. To help prefire planning efforts across California, areas that are most susceptible to postfire debris flows were identified before fire occurs. A calibration method was developed that relates existing vegetation type to fire severity, a critical input to the USGS’s postfire debris-flow likelihood model. Colleagues and I predicted debris-flow likelihood, volume, and combined hazard. We also created statewide maps that use simulated fire frequency and rainfall data to predict the likelihood that a basin will experience a wildfire and subsequent debris flow. We suggest that the model predictions are useful for identifying areas that pose the greatest risk of postfire debris-flow hazards. In addition, I compare these results to debris-flow likelihood results produced by the USGS for the Palisades and Eaton Fires, and I discuss future work to improve postfire geohazard predictions before wildfire.

March 12, 2025
noon - 1 p.m.
Geology 1707

Presented By:

• Hao Zhang - USC

Seminar Description coming soon.