Computational Math Seminar - Eileen Martin

Eileen Martin, Department of Geophysics, Colorado School of Mines

Fast algorithms for cross-correlation based analysis in seismology

Over the past two decades, a method called ambient noise interferometry has revolutionized subsurface seismic imaging for environmental, infrastructure and safety purposes. Ambient noise interferometry is based on cross-correlation of many time series to explore potential time-lagged relationships between them, often indicating an estimate of a Green's function in an area. Likewise, template matching (another cross-correlation based technique) has enabled huge growth of earthquake catalogs by detecting small events similar to larger known events. Both cross-correlation based techniques require significant data movement, and new seismic sensor technologies are leading to rapidly growing sensor arrays, resulting in major scalability challenges for seismic data analysis. In this talk I will show some new methods to calculate array-wide cross-correlations that take advantage of lossy data compression to reduce data movement and computational costs by performing cross-correlations directly on compressed data. Seismologists often use the cross-correlation results as an input to a few types of array beamforming methods (similar to beamforming used in wireless communications and astronomy). In fact, we can calculate these final beamforming results directly from the ambient seismic noise with new algorithms that never explicitly calculate cross-correlations and improve scalability from quadratic to linear.