Deep Learning‐Based <i>H‐κ</i> Method (HkNet) for Estimating Crustal Thickness and <i>Vp/Vs</i> Ratio From Receiver Functions

The teleseismic receiver function (RF) is commonly used to determine major interfaces of the Earth. If the crustal Vp is known approximately, the Moho converted Ps phase and crustal multiple reverberations can be used to determine the thickness (H) and average Vp/Vs ratio (κ) of the crust. A widely used method for this is H-κ stacking (Zhu & Kanamori, 2000, https://doi.org/10.1029/1999JB900322), which uses grid search and superposition to find the maximum coherent energy of the Moho Ps and its reverberated multiples phases. However, this method assumes a homogeneous isotropic crust and a flat Moho. Furthermore, it is affected by the reference crustal Vp. Improved methods, such as the H-κ-c method for anisotropic media and inclined interfaces (J. Li et al., 2019, https://doi.org/10.1029/2018JB016356), help alleviate the problem. In this paper, we propose a new method that uses deep learning to estimate H and κ. Our method is divided into two steps. The first step employs a denoise architecture (the DenoiseNet) to reduce the noise level of the RFs and restore missing back-azimuthal information. In the second step, our new deep learning network (the HkNet) is used to estimate H and κ. Deep learning has the inherent ability to automatically extract complex features from RFs, which allows us to estimate the parameters in complex media with different crustal Vp. Synthetic data tests show that the proposed method achieves better accuracy than the H-κ and H-κ-c methods. Applications to real data show that the proposed method is robust and reliable in a wide range of geological settings.