Reweighted Partial Least Squares and Deep Learning-Based RDOA/AOA Estimation for Seismic Epicenter

In this study, we proposed an innovative approach to accurately estimate the location of a seismic epicenter using a combination of reweighted partial least squares and deep learning-based range-difference-of-arrival and angle-of-arrival models. Our method enhances existing P- and S-wave estimation techniques by employing a short-time Fourier transform to analyze the seismic signals detected at the four nearest stations. The resulting spectrogram images were used to develop a deep learning model that accurately distinguishes P-waves, S-waves, and noise. By classifying images according to the time variation, we defined the new onset time with the highest probability as the P- and S-wave arrival times. The reweighted partial least squares method significantly improved the accuracy and robustness of the range-difference-of-arrival and angle-of-arrival models for epicenter localization. The proposed method demonstrated an improved epicenter localization accuracy in the simulation of real and ideal cases. The proposed process of epicenter localization is a potential solution for various seismic monitoring and early warning systems.