Determining Rupture Directivity of 2019 Ridgecrest sequence aftershocks with Gaussian Mixture modeling

Of the numerous earthquakes that occurred during the 2019 Ridgecrest sequence, only the mainshock and foreshock ruptures have been thoroughly analyzed. Given the existence of orthogonal faulting at many scales, we decided to investigate the rupture directivity of the smaller aftershocks to identify the most likely fault planes for each. To do this, we analyzed a large dataset of earthquake source spectra using a Gaussian Mixture Model (GMM). The GMM is a type of generative model that describes a dataset as a superposition of Gaussian distributions. Fitting a GMM entails inferring the latent variables that parameterize each Gaussian, where each Gaussian is a distinct mode. Here, we treat end-member rupture directivity modes as the GMM modes. We fit a GMM to apparent radiated energy values at each station for an ensemble of events. In total, we analyzed over 14,000 events across 40 stations. The resulting modal probability densities capture azimuth-dependent signals, and with a geographically-clustered approach, allow us to track how the rupture propagation probabilities evolve over the structurally complex Ridgecrest area.