Locating tremor using stacked products of correlations

We introduce a back-projection method to locate tremor sources using products of cross-correlation envelopes of time series between seismic stations. For a given subset of n stations, we calculate the (n − 1)th-order product of cross-correlation envelopes and we stack the back-projected products over combinations of station subsets. We show that compared to existing correlation methods and for realistic signal and noise characteristics, this way of combining information can significantly reduce the effects of correlated (spurious or irrelevant signals) and uncorrelated noise. Each back-projected product constitutes an individual localized estimate of the source locations, as opposed to a hyperbola for the existing correlation techniques, assuming a uniform velocity in two dimensions. We demonstrate the method with synthetic examples and a real-data example from tremor at Katla Volcano, Iceland, in July 2011. Despite very complex near-surface structure, including strong topography and thick ice cover, the method appears to produce robust estimates of tremor location.