Harnessing AI to uncover hidden earthquake signals, offering crucial insights months before disaster strikes.
Study: Abnormal low-magnitude seismicity preceding large-magnitude earthquakes. Image Credit: metamorworks / Shutterstock
In an article recently published in the journal Nature Communications, researchers comprehensively explored precursory signals of large-magnitude or potentially destructive earthquakes by identifying patterns in abnormal low-magnitude seismic activity that often precede major events. They used machine learning (ML) techniques to analyze earthquake catalogs to improve understanding of earthquake precursors and enhance seismic warning systems and forecasting capabilities.
Background
Forecasting earthquakes has long been a challenge in seismology. Traditional methods have examined various geophysical, geochemical, and biological precursors, such as thermal anomalies, acoustic emissions, changes in groundwater levels, and crustal deformation. Although recent studies suggest that strong warning signs may appear shortly before significant earthquakes, identifying these signs is challenging due to the complex patterns and variability of seismic activity.
Despite improvements in data collection and analysis, these indicators remain unreliable. However, recent advancements in artificial intelligence (AI) techniques such as ML and deep learning (DL) provide new ways to detect subtle, nonlinear patterns in seismic data that might signal large earthquakes. This study uses these advancements to explore the potential of low-magnitude seismicity as a reliable precursor to significant seismic events.
About the Research
In this paper, the authors investigated low-magnitude seismicity before two major earthquakes: the 2019 Ridgecrest sequence in California and the 2018 Anchorage earthquake in Alaska. They employed a supervised ML algorithm, specifically a random forest model, to analyze earthquake catalogs and identify abnormal seismic activity. This algorithm combines statistical features with finite element solid mechanics models to detect anomalies in low-magnitude seismicity.
The study focused on earthquakes with magnitudes between 1 and 6, examining their spatiotemporal distribution to find patterns that could indicate impending large-magnitude events. Historical seismic data from Southern California and Southcentral Alaska were used to train the algorithm. The model aimed to determine if the target events were preceded by anomalies in low-magnitude seismicity that could act as early warning signals.
Finite element solid mechanics models were also used to support the hypothesis that abnormal low-magnitude seismicity is linked to significant increases in pore fluid pressure within large fault segments. This pressure escalation causes uneven changes in the regional stress field, which the algorithm can detect as precursory signals.
Key Findings
The outcomes showed that both the 2019 Ridgecrest sequence and the 2018 Anchorage earthquake were preceded by up to three months of abnormal low-magnitude seismicity. This activity covered 15-25% of the region in Southern California, while it was similar in Southcentral Alaska. The ML algorithm successfully identified these anomalies, demonstrating its potential for real-time seismic monitoring.
The researchers observed a significant increase in the likelihood of a large-magnitude earthquake occurring within 30 days in the weeks leading up to the main events. For the Ridgecrest sequence, this probability rose to 75% about 40 days before the first earthquake and peaked at 90% after the initial shock. For the Anchorage earthquake, the probability reached 85% just days before the event.
The study also found that abnormal low-magnitude seismicity was not limited to the epicentral areas but spread across multiple fault zones. This widespread tectonic unrest suggests significant changes in the regional stress field as large fault segments approach failure. The authors proposed that these changes result from the softening of large faults due to increased fluid circulation, which reduces fault stiffness and causes uneven stress distribution.
Applications
This research has significant implications for seismic monitoring and early warning systems. By identifying abnormal low-magnitude seismicity as a reliable precursor to large earthquakes, the authors provided a valuable tool for predicting major seismic events weeks or months in advance. This could improve preparedness and reduce the impact on vulnerable communities.
The ML algorithm developed can be integrated into existing systems to detect tectonic unrest and issue timely warnings. It also shows promise for refining the predicted size and location of major earthquakes, enhancing forecast accuracy. Additionally, this method could be applied to other regions with comprehensive earthquake catalogs, potentially improving global forecasting capabilities.
Conclusion
In summary, ML algorithms proved effective for identifying precursory signals of large-magnitude earthquakes. The researchers found that regional tectonic unrest, indicated by abnormal low-magnitude seismicity, could serve as an early warning for major seismic events. This approach could enhance earthquake preparedness and reduce risks in seismically active regions worldwide.
Future work should focus on refining these methods and testing their applicability in various tectonic settings, contributing to more effective seismic warning systems. Overall, this research represents a significant advancement in understanding fault network dynamics and improving the reliability of seismic forecasts.