University of Alaska Fairbanks: They developed an algorithm that can predict earthquakes

The University of Alaska Fairbanks (UAF) published revolutionary research in Nature Communications journal that suggests that they can predict future earthquakes with the aid of machine learning. Their theory is based on the facts that they summarized at two recent strong earthquakes, the 7.1 magnitude earthquake in 2018 at Anchorage and the 7.1 magnitude earthquake in 2019 at California. The researchers noticed that prior to the earthquakes at the area of the epicenter unusual seismic activity and patterns took place that ultimately led to the seismic events.

Specifically, Társilo Girona, who is research assistant professor at UAF alongside with Kyriaki Drymoni from Ludwig-Maximilians-Universität in Munich, claims that they created models that after analyzing seismic data, they can identify unusual low-magnitude earthquake activity. They argue that this behavior often indicates a strong earthquake would take place as they use advanced statistical methods, mainly machine learning to analyze data from earthquake catalogs.

Among their findings was that the algorithm they developed identified unusual seismic activity in 15% to 25% of the affected areas, along with multiple tremors with magnitudes 1.5 or less about three months before each earthquake. In addition, when they analyzed the data for the Anchorage earthquake, they found that the algorithm gave about 80% probability that a strong earthquake will strike a month before while, that probability rose to 85% the week before.

The researchers also argue that the seismic events might be connected to increased pore fluid pressure within faults, as they claim that this abnormal pressure can result in the change of the mechanical properties of the faults.  Therefore, this is why they found the abnormal low-magnitude earthquake activity. Their study can have significant impact to the research community as the ability to predict major earthquakes can help in disaster preparedness as they can enable on time evacuations from high-risk areas, they can advise for reinforcement of critical infrastructure, decrease of economic losses through preventive measures while emergency services can be prepared.

In all, the research team is moderately optimistic about their findings, however, they know that the method they developed requires more testing before it will apply in real- situations, while the algorithm must be trained further on historical seismic data specific to each region where it might be implemented. In addition, they argue that predicting an earthquake is very complex because on the one hand an accurate forecast could save lives, on the other hand a false alarm could cause panic and the economy could easily be disrupted, while if they fail to predict the event, could lead to a disaster. Therefore, they believe that before their method is used in real life, policymakers and scientists must collaborate to establish guidelines for its responsible use.