A groundbreaking study by a scientist at the University of Alaska Fairbanks reveals that a new AI model can predict major earthquakes several months before they occur. The research, published in Nature Communications, highlights the potential of using machine learning to detect early signs of seismic activity and provide advance warnings to the public.
The study, led by Assistant Research Professor Társilo Girona from the Geophysical Institute at the University of Alaska Fairbanks, focuses on identifying large-magnitude earthquakes by analyzing seismic catalogs. Girona, an earth scientist and data scientist specializing in volcanic eruptions and seismic precursors, collaborates with geologist Kyriaki Drymoni from the Ludwig Maximilian University of Munich.
The study analyzed two major earthquakes: the 7.1 Mw Anchorage earthquake in 2018 and the 6.4 to 7.1 Mw Ridgecrest earthquake sequence in California in 2019. They found that in the months leading up to these earthquakes, approximately 15% to 25% of the regions in Alaska’s south-central and southern California experienced abnormal low-magnitude seismic activity.
In the Anchorage earthquake, the probability of a large earthquake occurring within 30 days or less jumped to about 80% around three months before the event. Just days before the earthquake, the probability increased to about 85%. Similar patterns were observed for the Ridgecrest earthquake sequence.
The researchers attributed the low-magnitude precursor activity to the increase in pore fluid pressure within fault zones. Pore fluid pressure refers to the pressure of fluids within rocks. When pore fluid pressure is high enough to overcome the frictional resistance between rock blocks on either side of a fault, it can lead to fault slip.
The increase in pore fluid pressure in faults that cause large earthquakes alters the mechanical properties of the fault, leading to uneven changes in the regional stress field. These changes are believed to control the abnormal, precursory low-magnitude seismic activity.
Girona and Drymoni’s research emphasizes the potential of machine learning to identify meaningful patterns that may indicate an impending earthquake. Modern seismic networks generate vast datasets that, when analyzed properly, can provide valuable insights into seismic events.
The researchers developed a computer algorithm to search for abnormal seismic activity in the datasets. The algorithm is a set of computer instructions that guide a program to interpret data, learn from it, and make informed predictions or decisions.
Machine learning is having a significant positive impact on earthquake research, Girona said. Modern seismic networks generate massive datasets that can provide valuable insights into seismic precursors if analyzed properly. This is where the advancements in machine learning and high-performance computing can play a transformative role, allowing researchers to identify meaningful patterns that may indicate an impending earthquake.
The authors note that their algorithm will be tested in near-real-time scenarios to identify and address potential challenges in earthquake forecasting. They emphasize that the method should not be used in new areas without being trained on the region’s historical earthquake data.
Accurate earthquake forecasting has significant ethical and practical implications. While accurate forecasts can save lives and reduce economic losses by providing advance warnings for timely evacuations and preparations, the inherent uncertainty of earthquake forecasting also poses major challenges. False alarms can lead to unnecessary panic, economic chaos, and a loss of public trust, while forecasting errors can have catastrophic consequences.
The study represents a significant step forward in earthquake forecasting and highlights the potential of machine learning to improve our ability to predict and prepare for major earthquakes. As the research continues to evolve, it is hoped that this technology will ultimately contribute to a safer and more resilient future for communities around the world.
Views: 0