Array analysis and precise source location of deep tremor in Cascadia

<jats:p>We describe a new method to estimate the <jats:italic>S</jats:italic>‐<jats:italic>P</jats:italic> time of tremor‐like signals and its application to the nonvolcanic tremor recorded in July 2004 by three dense arrays in Cascadia. The cross correlation between vertical and horizontal components indicates that very often the high‐amplitude tremor signal contains sequences of <jats:italic>P</jats:italic> and <jats:italic>S</jats:italic> waves characterized by constant <jats:italic>S</jats:italic>‐<jats:italic>P</jats:italic> times (<jats:italic>T</jats:italic><jats:sub><jats:italic>S</jats:italic>‐<jats:italic>P</jats:italic></jats:sub>) in the range 3.5–7 s. A detailed observation of the three component seismograms stacked over the array stations confirms the presence of <jats:italic>P</jats:italic> and <jats:italic>S</jats:italic> wave sequences. The knowledge of the <jats:italic>T</jats:italic><jats:sub><jats:italic>S</jats:italic>‐<jats:italic>P</jats:italic></jats:sub> poses a strong constrain on the source‐array distance, which dramatically reduces the uncertainty on source locations when used with more traditional array processing techniques. Data were analyzed using the zero lag cross‐correlation technique (ZLCC) to estimate the propagation properties of the most correlated phases in the wavefield. Detailed polarization analyses were computed using the covariance matrix method in the time domain. Polarization parameters, joint with the results of ZLCC, allows for the discrimination between <jats:italic>P</jats:italic> and <jats:italic>S</jats:italic> coherent waves. Results show that the tremor wavefield is composed mostly by shear waves, although a consistent amount of coherent <jats:italic>P</jats:italic> waves is often observable. The comparison of the back azimuth at the three arrays indicate that the source of deep tremor migrates over a wide area, and often many independent sources located far from each other are active at the same time. The tremor source was located by a probabilistic method that uses the results of ZLCC, given a velocity model. When available, the inclusion of the <jats:italic>T</jats:italic><jats:sub><jats:italic>S</jats:italic>‐<jats:italic>P</jats:italic></jats:sub> time in the location procedure strongly reduces the depth range, with a distribution of hypocenters very near the subduction interface. This result, significantly different compared with previous less precise locations, makes the Cascadia nonvolcanic tremor more similar to the nonvolcanic tremor recorded in Japan, at least in cases of measurable <jats:italic>T</jats:italic><jats:sub><jats:italic>S</jats:italic>‐<jats:italic>P</jats:italic></jats:sub>. The polarization azimuth aligned with the slow slip direction and the source located on the plate interface indicate that deep tremor and slow slip are two different manifestations of a common phenomenon related with the subduction dynamics.</jats:p>