Assessing the predictive power of GPS-based ground deformation data for aftershock forecasting

We present a machine learning approach for aftershock forecasting of the Japanese earthquakes catalog. Our method takes as sole input the ground surface deformation as measured by Global Positioning System (GPS) stations on the day of the mainshock to predict aftershock location. The quality of data heavily relies on the density of GPS stations: the predictive power is lost when the mainshocks occur far from measurement stations, as in offshore regions. Despite this fact and the small number of samples and the large number of parameters, we are able to limit overfitting, which shows that this new approach is very promising.