Machine Learning, Robotic Mapping, and Particle Dynamics for Earthquake Geology

Abstract:  Documenting and interpreting the interaction between surface and tectonic processes for records of earthquakes in the landscape is a central challenge for both fundamental understanding and application to hazard assessment. Advances in the automation and scaling of data collection and data analysis for earth sciences are evident from robotics and artificial intelligence (or machine learning -- ML).  Central to the practice of robotic science is end-to-end simulation.  Robotic design, development, and simulation suites are based on the Robotic Operating System, are implemented in virtual environments (with capability for ML; e.g. Gazebo), and can employ powerful physics engines (e.g., Bullet).  In this talk, I will illustrate the potential for these tools in advancing earthquake geology including searching for, mapping, and analyzing the dynamics of fragile geologic features.  Measuring rock fragility can be challenging at scale.  Our work on ML-based mapping of rock traits (size, orientation, ellipticity) of ~240k particles on rocky fault scarps provides a new view for interpreting process and the effect of changing slip rate.  Once particles are identified, we move into their 3D mapping.  These 3D objects may precarious and thus negative indicators of strong ground motion.  Interpreting their stability may be done using our "virtual shake robot".  The VSR allows for exploration of the effects of input ground motion histories on the fragility of arbitrarily complex objects and substrates with realistic physics.  These tools present an important new set of capabilities for earthquake geology and beyond.

Bio:  Ramon Arrowsmith has been at Arizona State University since 1995 where he teaches field geology, structural geology, geomorphology, and computers in earth and space exploration. He has more than 30 years’ experience studying the earthquake geology, paleoseismology, and geomorphology of fault zones and publishing about their history of activity and hazards. He is cofounder and co-PI of the OpenTopography effort—a portal to the largest collection of freely available high resolution topography data (http://www.opentopography.org/).

He has led many short courses, workshops, and visioning activities emphasizing high resolution topographic data and tectonic geomorphology. He and his students research on active faulting, earthquake geology, tectonic geomorphology, and the geologic framework for human origins. He has held a number of administrative positions in the ASU School of Earth and Space Exploration (SESE) including associate chair (of Geological Sciences Department), associate directors of graduate studies and of operations, and is currently the associate director of operations for SESE.