Earthquake Cycle Models

Earthquakes are not isolated phenomena, but rather they are a small and destructive snapshot over a much longer seismic cycle. Understanding the seismic cycle can lead to a clearer understanding of earthquake occurrence and the associated seismic hazard in an area. In this particular area of research, I utilize numerical methods to generate long term synthetic earthquake catalogs, explore those catalogs and try to relate the model results to real life hazard.

Quasi-static Earthquake Cycle Modeling of Faults in the California East Bay (Ongoing)

In this Pacific Gas and Electric funded research I utilize the quasi-static physics based earthquake simulator RSQSim (Richards-Dinger & Dieterich, 2012) to run a suite of earthquake cycle models for a 5 segment fault system for the California East Bay based on the Statewide California Earthquake Center Community Fault Model V 7.0 (Plesch et al., 2024). We focus our attention on the Hayward, Calaveras, Rodgers Creek, Napa and Franklin faults. We run models with different fault friction, connectivity and coseismic weakening. Initial results show the sensitivity of multi-fault ruptures to the connectivity and cosiesmic weakening. In addition, our models can produce time-independent and dependent 30-year probabilities consistent with the Uniform California Earthquake Rupture Forecast Version 3 (Field et al., 2015;2015).

Figures from the RSQSim generated earthquake catalogs for the California East Bay.  On the left is an example of a Mw 7.35 earthquake from the catalog that ruptures both the Hayward and Rodgers Creek faults. On the right hand side are the Time-Dependent comparisons with UCERF3 (top) and the number of events that nucleate on a segment and rupture multiple segments (bottom).

Machine Learning Classification Driven by Earthquake Simulations (Niyogi, Ghosh, Marschall, et al. 2025 JGR)

In this study we utilized RSQSim to generate a synthetic earthquake catalog for the San Andreas and Garlock fault intersection in California. We classified each rupture based on which fault the earthquake nucleated on and how many segments it ruptured. For each classified earthquake in the catalog, we extracted an average of the initial on fault parameters and fed the results into several machine learning algorithms. We demonstrate that for certain scenarios the machine learning models have considerable testing accuracy in classifying the rupture path, and augmentation of features in the input training data leads to improvements in the accuracy both in terms of precision and recall. Furthermore, our machine learning models suggest that the pre-earthquake conditions of the fault on which the rupture nucleated are the dominant parameters which affect rupture path at the branch intersection.  

Figures from Niyogi et al., 2025 (JGR) Showing the faults used in the study on the left and machine learning output cumulative importance scores for the top five parameters utilized by the machine learning algorithm in its prediction of classifying the rupture path.