Speaker
Description
We present results of testing the latest version of our Probabilistic Event Detection, Association, and Location (PEDAL) algorithm. PEDAL uses an Earth model discretized into a dense 3D grid of 427,265 nodes, extended to 4D by the addition of a time dimension. Given a set of seismic observations (arrival time, horizontal slowness, azimuth and associated uncertainties), a 'fitness’ value is calculated at each grid node, assuming that each observation was generated by a refracted P wave. The node with peak fitness value is accepted as a hypothetical seismic event location. We then solve for the corresponding origin time and associate individual arrivals with the event, considering many different phases. Improvements include: 1) incorporating prior probability of signal detection for each station; 2) association in two stages, P first, then later phases; 3) calculation of mb and association based on magnitude, depth, and distance from event to station; and 4) integration with waveform correlation. We tested the new version on a 2-week period of time processed by the IDC and carefully examined by an analyst to identify all legitimate events. A sophisticated bulletin review algorithm shows PEDAL performance superior to the Global Associator (GA).
