SCEC Project Details
SCEC Award Number | 24186 | View PDF | |||||||
Proposal Category | Collaborative Research Project (Multiple Investigators / Institutions) | ||||||||
Proposal Title | Knowledge Informed Neural Temporal Point Process for Earthquake Forecast | ||||||||
Investigator(s) |
|
||||||||
SCEC Milestones | D1-1, D2-2, D3-1 | SCEC Groups | EFP, Seismology | ||||||
Report Due Date | 03/15/2025 | Date Report Submitted | 04/06/2025 |
Project Abstract |
In recent years, notable progress has been made in earthquake forecasting, particularly with the development of deep learning models that integrate temporal point processes. However, one of the primary limitations of these models is their focus on forecasting seismicity rates which are dominated by the smallest events that have little impact on hazard. The Epidemic Type Aftershock Sequence (ETAS) model can model the distribution of seismic events in both time and space. While ETAS provides valuable insight into the temporal and spatial clustering of aftershocks, it fails to address an important aspect of earthquake forecasting: the magnitude of future events. To address this, we investigate a novel AI model, referred to as the Spatio-Temporal Earthquake Forecasting with Informed Point-process Optimization (STEF-IPO), which combines the strengths of both neural networks and temporal point processes. Our experiment analysis shows that the proposed STEF-IPO model achieved an Area Under the Receiver Operating Characteristic Curve (AUC-ROC) of 78.67% for predicting events with a magnitude greater than 4.5. |
Intellectual Merit |
Earthquakes are a prime example of a complex natural process with far-from-equilibrium nonlinear dynamics that is not well understood. The project has advanced significantly the understanding of the limits to forecasting of earthquakes, dynamics of seismicity, and processes that characterize time intervals preceding large events, using novel AI techniques combined with state-of-the-art domain knowledge including results from statistical seismology, earthquake physics, and various data sets. Innovative contributions have been made to domain-informed deep learning by constructing and integrating domain knowledge-based graphs. |
Broader Impacts | Earthquakes are the second most devastating natural disaster causing human fatalities, injuries and substantial financial losses. About half of the worldwide fatalities from natural disasters in the last 40 years were caused by earthquakes and associated cascading hazards such as landslides, fires and tsunamis. An improved ability to forecast large earthquakes will reduce the vulnerability of major metropolitan areas near large hazardous faults in the US (e.g., all major cities in California) and worldwide. |
Project Participants |
Yan Liu, Yehuda Ben-Zion, Yizhou Zhang, Zijun Cui Collaborator: Mizrahi Leila |
Exemplary Figure | Figure 1 |
Linked Publications
Add missing publication or edit citation shown. Enter the SCEC project ID to link publication. |
|