Deep learning-based H- method (HKNet): A new method for estimating crustal thickness and Vp/Vs ratio from receiver functions

The teleseismic receiver function is commonly used to image major interfaces of the Earth, which are important for understanding regional geology and tectonic evolution. If the average P-wave velocity in the crust is known, then the Moho converted Pms phase and crustal multiple reverberations (PpPms and PpSms+PsPms) in the receiver function can be used to determine the thickness (H) and average Vp/Vs ratio () of the crust. A widely used method is the H- stacking method (Zhu LP and Kanamori, 2000), which uses grid search and superposition to find maximum coherent energy of the Pms phase and the multiples. However, the method assumes a homogeneous isotropic crust and a flat Moho interface. Furthermore, it is affected by the reference P-wave velocity and intra-crustal discontinuities. Improved methods, such as the H--c method for anisotropic media and inclined interfaces (Li JT et al., 2019), help alleviate the problem. In this study, we explore a new method based on deep learning to estimate crustal H and (named as HKNet). The H and estimation problem can be regarded as a seismic phase picking problem for the crustal converted and multiple phases. Deep learning has the inherent ability to automatically extract complex features in the receiver functions, which allows us to estimate the parameters in complex media. Information from both radial (R) and transverse (T) components can be equally used, which helps greatly in constraining the parameters. Our training data set is composed of synthetic receiver functions for models of different levels of complexity. The H and values in the test model are treated as the labels and the corresponding R and T components as a function of azimuth are treated as 2D data (images). Synthetic tests show the new method achieves much better accuracy over the H- and H--c methods. We will show applications to real data. References Zhu L , Kanamori H . (2000). Moho depth variation in southern California from teleseismic receiver functions. J. Geophys. Res., 105(B2),2969-29890. Li JT , Song XD, Wang P , Zhu LP. ( 2019). A Generalized H- Method With Harmonic Corrections on Ps and Its Crustal Multiples in Receiver Functions. J. Geophys. Res., 2019, 124(16).