East Asia Regionalization Based on Receiver Functions
Abstract:We present here a first step toward a model of the seismic velocity structure through stochastic full waveform tomography of East Asia. Such inversion typically requires exploring thousands of variations of each parameter that is inverted to ensure reliable probabilistic inferences. Here we explore regionalization in order to reduce the dimension of the parameter space and identify regions with similar seismic characteristics that can be treated as a common block in the subsequent stochastic inversion. We follow the approach of Lekic and Romanowicz (2011) in which regionalization is performed through a cluster analysis of tomographic velocity models. Our analysis is based on teleseismic P-wave Receiver Functions (RFTNs) instead. We apply a K-means algorithm minimizing a distance metrics defined in the vector space of RFTNs. Different metrics have been tested to optimize the clustering. Coherence and association with known tectonic and physiographic features and/or established geophysical information is also tested.
We first validate our clustering analysis with two different receiver functions datasets from USArray stations. A first set was built by stacking EARS receiver functions for each available station across all azimuths and with a Gaussian filter width of 2.5 Hz and small ray parameters (0.038 to 0.05 s/km). The second dataset was built by interpolation of the receiver function wavefield (Chai et al, 2015). The agreement with geological and tectonic regions of western US is better for the latter dataset.
Our future inversion region is East Asia where remarkable instrumentation efforts open the door to high-resolution tomography studies at the continental scale. Our collaborators provided teleseismic P-wave receiver functions for 785 stations. Clustering results show agreement with known tectonic features of the area, and we will show our attempts of regionalization based on this information.