2024 Volume 24 Issue 5 Pages 5_35-5_44
Recently, PoDIM (POsition-dependent Deep Inpainting Method), which is based on a deep inpainting method, has been proposed to obtain spatially continuous seismic motion data from sparse observations. PoDIM utilizes a positional feature map that represents the degree of amplification or attenuation at each position to realize position dependent interpolation processes. However, since the maps are obtained through the training of a complex deep model starting with random values, it was difficult to interpret their roles. Therefore, in this study, we propose a position-dependent interpolation that can generate and interpret position feature maps based on top surface depth data for multiple s-wave velocities that directly affect the propagation of seismic motions. The effectiveness of the proposed method is then demonstrated through experiments using simulated data of a hypothetical Nankai Trough earthquake.
