We present a probabilistic model of seismic phases whereby the
attributes of the body-wave phases are correlated to those of the first
arriving P phase. This model has been incorporated into NET-VISA
(Network processing Vertically Integrated Seismic Analysis) a
probabilistic generative model of seismic events, their transmission,
and detection on a global seismic network. In the earlier version of
NET-VISA, seismic phase were assumed to be independent of each
other. Although this didn't affect the quality of the inferred seismic
bulletin, for the most part, it did result in a few instances of
anomalous phase association. For example, an S phase with a smaller
slowness than the corresponding P phase.
We demonstrate that the phase attributes are indeed highly correlated,
for example the uncertainty in the S phase travel time is significantly
reduced given the P phase travel time. Our new model exploits these
correlations to produce better calibrated probabilities for the events,
as well as fewer anomalous associations. |