Scalable P-wave passive seismic topography


The goal of this work is to simultaneously reconstruct the location-dependent velocity field of seismic P-waves in an underground mine, and the unknown seismic sources during a given period, or dynamically over time, by using only registered data of passively generated P-waves (in particular, not relying on calibration shots as is usually the case). To that end, we develop a two-step algorithm based on Bayesian modeling and on the use, to our knowledge for the first time in geostatistics applications, of a stochastic gradient descent method, a scalable approach which has proved extremely successful in recent developments in Machine Learning.  This method allows our algorithm to  deal with the heavy amount of historical data from the world’s largest underground mine, while gaining increasing accuracy with respect to the method employed on-site. We present a proof of concept on synthetic data, and first results on real data in a simplified setting. Perspectives will also be discussed.  Based on work in progress with J. Fontbona, D. Neira and J. Prado (CMM).

Date: Jul 03, 2018 at 15:00 h
Venue: Sala AMTC 101, Beauchef 851, torre Norte.
Speaker: Claire Delplancke
Affiliation: Postdoctoral Researcher, Center for Mathematical Modeling
Coordinator: Prof. Joaquín Fontbona

Posted on Jul 4, 2018 in Seminario CMM-Data & CMM-Mining, Seminars