We characterize the source of an earthquake based on identifying the nodal lines of the radiation pattern it produces. These characteristics are the mode of failure of the rock (shear or tensile), the orientation of the fault plane and direction of slip. We will also derive a correlation coefficient comparing the source mechanisms of different earthquakes. The problem is formulated in terms of a simple binary classification on the surface of the sphere. Our design goal was to derive an algorithm that would be both robust to misclassification of the observed data and suitable for online processing. We will then go on to derive a mapping which translates the learned solution for the separating hyper-plane back to the physics of the problem, that is, the probable source type and orientation. For reproducibility, we will demonstrate our algorithm using the example data provided with the HASH earthquake classification software, which is available online.

Keywords:machine learningearthquakehazardclassification.