Oral Presentation

Machine learning has revolutionized data processing, yet collecting large datasets for its application in earthquake science remains a challenge. This talk introduces the use of robotics to automate data collection, a crucial step toward automating and advancing geoscience research. I present three cases where integrating robotics and machine learning opens new avenues for fault zone characterization and fragile geological feature analysis. In bedrock fault zones, understanding the geometric distributions of rocks is essential for elucidating fault scarp formation and development processes. I have employed UAVs and deep-learning techniques for large-scale, high-resolution analysis of rocky fault scarps in the California Volcanic Tableland, enabling the detection and analysis of a large number of diverse rock geometries and distributions. Strong correlations between rock geometries and geomorphic features inform a particle transportation model, providing a new perspective that extends past research on macroscopic scarp geometry such as slope, height, and length. In new work, I am designing a multi-UAV system to autonomously collect and map millimeter-scale fracture data, offering rapid earthquake damage assessment. Following an earthquake, a scout UAV autonomously locates surface faulting perimeters, and a fleet of low-flying mapping UAVs captures high-resolution images, which are then analyzed by deep neural networks to extract detailed fracture geometry and surface deformation characteristics. Finally, for fragile geological features such as precariously balanced rocks (PBRs), I have developed methods using UAVs and deep learning for automated detection, mapping, and dynamic analysis. Utilizing a virtual shake robot powered by a physics engine, I analyze PBR overturning and large-displacement dynamics, providing ground motion constraints essential for earthquake studies. These advances illustrate how improved methods for dense data collection can provide new insights into earthquake surface rupture and ground motions, advancing our understanding of the impacts of earthquakes on the landscape.