Project:	Analysis of Urban Seismic Signals Using Machine Learning Techniques
Disciplines:	Computer Science, Data Science
Mentor:	Monica Kohler, Research Professor of Civil Engineering, (EAS), kohler@caltech.edu, Phone: 626-395-4142
Mentor URL:	kohler.caltech.edu  (opens in new window)
Background:	At Caltech we have developed a novel seismic network (Community Seismic Network) that includes the permanent installation of over 1000 seismic sensors around the greater Los Angeles area for earthquake monitoring. The sensors are triaxial accelerometers that record waveform time series data 24/7/365; thus, they record both local earthquakes as well as shaking due to non-earthquake shaking (“ambient vibrations”) events such as storms, construction, vehicular traffic, motors, and other anthropogenic activities. It is essential to be able to characterize these sources of ambient vibrations in order to characterize their patterns and significance, and to be able to distinguish them from natural hazard events such as earthquakes. The >1000 seismic sensors distributed around the city have recorded many months of continuous data, so there is a large, rich dataset available for analysis. This SURF project will involve analysis of that dataset, taking advantage of machine learning tools that are applicable for constructing models of the data. The models produced by the SURF will be investigated to see if they can predict temporal and spatial patterns in the non-seismic vibrations, and if they illustrate patterns in the data that are not obvious from application of traditional methods. The goals are to quantify how the underlying geology of the Los Angeles basin affects the background noise levels in the seismic signals, and to discover the human-generated and weather-related sources of the non-earthquake signals.
Description:	The SURF student will apply machine learning tools to a large seismic dataset that has already been computed from the ambient vibration and earthquake acceleration records produced by the 1000-station Community Seismic Network. The pre-computed data include peak ground accelerations, L1-norms, and RMS accelerations, recorded for successive one-hour time windows. They also include power spectral density values computed on hour-long datasets. The student will apply machine learning algorithms to construct models of these parameters, and to identify correlations between the parameters and potential sources of vibrations. The tools may involve a combination of analyses that include clustering algorithms, active learning techniques, and specialized data labeling tools to gradually identify key features and categories. Both supervised and unsupervised learning techniques will be explored to identify potential features. The project is entirely computational, and the student will interact with the Community Seismic Network research and operations team at Caltech.
References:	1. Community Seismic Network website: http://csn.caltech.edu 2. Mohammed, S., R. Shams, C. C. Nweke, T. E. Buckreis, M. D. Kohler, Y. Bozorgnia, and J. P. Stewart, Usability of Community Seismic Network recordings for ground motion modeling, Earthquake Spectra, doi:10.1177/87552930241267749, 2024. 3. Clayton, R., M. Kohler, R. Guy, J. Bunn, T. Heaton, and M. Chandy, CSN/LAUSD network: A dense accelerometer network in Los Angeles schools, Seis. Res. Lett., 91(2A), 622-630, doi:10.1785/0220190200, 2020.
Student Requirements:	Linux and Python. Some familiarity with supervised and unsupervised learning techniques. Domain knowledge of seismology or civil engineering is not necessary.
Programs:	This AO can be done under the following programs:   Program    Available To        SURF    Caltech students only  Click on a program name for program info and application requirements.