A Bayesian Method to Apply the Results of Multiple-Event Seismic Location to a Subsequent Event

Abstract:

BayesLoc is a Bayesian multiple-event seismic locator that uses a Markov chain Monte Carlo (MCMC) algorithm to sample possible seismic hypocenters, travel-time corrections, and the precision of observed arrival data (absolute picks and differential times based on cross-correlated waveforms). By simultaneously locating multiple seismic events, regional biases in the assumed travel-time model (e.g., ak135) can be estimated and corrected for, and data from different seismic stations and phases can be weighted to reflect their accuracy/precision for an event cluster. As such, multiple-event locators generally yield more accurate locations than single-event locators, which lack the data to resolve the underlying travel-time model and adaptively “weight” the arrival data differently for each station and phase. On the other hand, single-event locators are computationally more attractive, making them more suitable for rapid (realtime) location of seismic activity. We present a novel approach to approximate the location accuracy of the BayesLoc multiple-event analysis at a computational cost that is comparable to BayesLoc single-event analysis. The proposed approach consists of two steps: a precomputed multiple-event training analysis and subsequent real-time, single-event location for new events. The precomputed training analsysis consists of carrying out a multiple-event BayesLoc run in a given target event cluster, yielding a posterior sample of travel-time corrections and weights. Given a new event in the vicinity of the training cluster, a BayesLoc single-event run is carried out which samples the travel-time corrections and weights from the multiple-event training run. Hence, it has all the benefits of the multiple-event run at the cost of a single-event run. We present the theoretical underpinnings of the new approach and we compare event location results for the full multiple-event, single-event, and the new approaches. This work was performed under the auspices of the U.S. Department of Energy by Lawrence Livermore National Laboratory under Contract DE-AC52-07NA27344. (LLNL-ABS-658134)