A Bayesian Lasso Method to Predict the Likelihood of Regional Seismic Phase Blockage

Abstract:

A common problem in monitoring seismology is that the seismic phases used to discriminate between different types of sources can be blocked due to path effects. Widespread regional phase (i.e., Sn and Lg) blockage is often the result of high attenuation in tectonically active regions, so mapping out blockage zones can reveal regions of melt in the uppermost mantle and is useful for seismic discrimination studies by predicting the likelihood of blockage. Seismic wave regional phase blockage is spatially systematic and correlates fairly strongly with low Q zones. Systematically blocked data would undermine any ability to correctly predict regional phase amplitude. In this study, we have developed a method to adapt logical tomographic techniques to predict the likelihood of a particular phase being blocked by applying a Bayesian Lasso approach to observations of blockage for the Sn phase. We have chosen Sn because it suffers from widespread blockage for continental paths more than any other regional phase. We have applied this method on simulated Sn blockage data sets with different blockage structures and on real data from the Middle East (ME) and China. To produce the simulated data, we used the real geometry of the ME and set the efficiency level by ray location and by the amplitude reduction. As illustrated by the confusion matrix, we see high accuracy for both simulated and real data prediction. This method enables us to generate a model that can predict the probability of phase blockage and also the uncertainly of that probability estimate. This is necessary because there are some regions where observations of blockage are not always consistent (i.e., there is noise in the blockage data).