Understanding the shape of faults is important for predicting how earthquakes behave and for improving earthquake hazard models. However, because faults are mostly hidden from view, they are often poorly understood. In recent years, better earthquake detection has made it possible to track more small earthquakes, giving us new ways to study faults beneath Earth's surface. We developed a Python software tool that automatically builds 3D models of faults using the locations of earthquakes. The tool first groups nearby earthquakes, then combines similar groups to find larger fault patterns. It uses machine learning to fit smooth surfaces through these groups and outputs the results as 3D fault models. It also calculates how tightly the earthquakes are clustered around each surface and how flat each surface is.

We tested this method in two complex fault areas in California: the junction of the San Andreas and Calaveras faults, and the site of the 2019 Ridgecrest earthquake sequence. Our results closely match expert-built models, showing that our tool can reliably capture complex fault shapes using only earthquake data. This makes it easier to create detailed 3D fault maps with minimal manual input.