Regional monitoring of infrasound events using multiple arrays: application to Utah and Washington State

In this paper, we present an integrated set of algorithms for the automatic detection, association, and location of low-frequency acoustic events using regional networks of infrasound arrays. Here, low-frequency acoustic events are characterized by transient signals, which may arise from a range of natural and anthropogenic sources, examples of which include (but are not limited to) earthquakes, volcanic eruptions, explosions, rockets and bolides. First, we outline a new technique for detecting infrasound signals that works successfully in the presence of correlated noise. We use an F-statistic, sequentially adapted to ambient noise conditions, in order to obtain detections at a given statistical significance while accounting for real background noise. At each array, individual arrivals are then grouped together based on measured delay-times and backazimuths. Each signal is identified as either a first or later arrival. First arrivals at spatially separated arrays are then associated using a grid-search method to form events. Preliminary event locations are calculated from the geographic means and spreads of grid nodes associated with each event. We apply the technique to regional infrasound networks in Utah and Washington State. In Utah, over a period of approximately 1 month, we obtain a total of 276 events recorded at three arrays in a geographic region of 6 x 4 degrees. For four ground-truth explosions in Utah, the automatic algorithm detects, associates, and locates the events within an average offset of 5.4 km to the actual explosion locations. In Washington State, the algorithm locates numerous events that are associated with a large coalmine in Centralia, Washington. An example mining-explosion from Centralia is located within 8.2 km of the mine. The methodology and results presented here provide an initial framework for assessing the capability of infrasound networks for regional infrasound monitoring, in particular by quantifying detection thresholds and localization errors.