Seismic tomography of Shatsky Rise by adaptive importance sampling

[1] We present results from a wide-angle seismic refraction survey over Shatsky Rise, a large oceanic plateau in the northwestern Pacific. A new Monte Carlo sampling scheme is developed to explore comprehensively the model space of crustal velocity structure by joint refraction and reflection seismic tomography. The new scheme, which is founded on the notion of adaptive importance sampling, is made possible by the combination of several independent developments, including the introduction of effective model parameters, the implementation of automated model regularization, and a polynomial expansion of the probability density function. From 21 ocean-bottom instruments deployed along a 420-km-long refraction transect, we collect 16446 Pg and 4735 PmP travel times, which are inverted to construct a two-dimensional compressional velocity model across the major axis of Shatsky Rise. The convergence of Monte Carlo sampling is tested by running parallel sampling chains, and the effective dimensions of the model space are estimated to be 10 by principal component analysis. The thickness of the rise crust varies from 9 km to 30 km, and the average velocity of the lower crust ranges from 7.0 km s 1 to 7.2 km s . One standard deviation uncertainty for whole-crustal thickness and average lower-crustal velocity is <1 km and 0.05 km s , respectively, for most of the model domain. Shatsky Rise crustal structure exhibits a negative correlation between crustal thickness and velocity, which implies that a chemically anomalous mantle may have been responsible for the formation of the rise.