56 thousand dead, 2 million displaced. These numbers shocked the world in 2023, when the 7.8 magnitude Turkiye-Syria earthquake violently rocked the volatile East Anatolian Transform Fault. Despite decades of data collection and geologic advancements, earthquakes continue to perplex scientists; the natural disaster’s unpredictable nature and destructive, fatal consequences for human life continue to haunt millions living in seismologically-active regions across the world. Every second counts in alert systems; just five extra seconds allows a surgeon to put down his scalpel, or for a pedestrian to escape the path of a falling utility pole. But what if cities had more than just five seconds? What if they had 24 hours, or, better yet, a full week in anticipatory warnings? Machine learning (ML) may transform this wishful thinking into reality. 

As a specific subset of artificial intelligence (AI), ML operates by studying vast amounts of data, recognizing common patterns, and independently refining itself to produce semi-accurate future projections. Tectonic ML algorithms commonly study data collected by seismometers, or small devices that record seismic waves (a wave-like series of vibrations produced by a release of energy). These seismometers vary in type; microelectromechanical systems (MEMS) and distributed acoustic sensing (DAS) systems measure particle movement and force, to name a few. This data is transported to ML algorithms, where it is then transformed into earthquake-incidence predictions; these predictions are near-instantaneously carried to alert systems via an Internet of Things (IoT) system. An IoT system, in simple terms, allows for the seamless exchange of data between devices, which is especially valuable in situations requiring real-time data collection and alert notifications. 

This modern intersection of ML and seismology has yielded promising results in recent years. The Los Alamos National Laboratory, for example, studied the Hawaiian Volcano Observatory over the course of approximately two months in 2018, during which 50 earthquakes occurred. As part of their research, the team took advantage of a global navigation satellite system (GNSS), known as GPS in the U.S., which is a popular method of measuring an object’s geopositioning through signals from satellites in space. The Los Alamos team utilized GNSS to measure minute displacements in the ground’s elevation; through an ML algorithm, they were able to successfully predict the next time the region’s stick-slip fault, a fault characterized by high stress and friction, would fail. They discovered a 30-second window of seismological data that clued towards the occurrence of a quake. This window, referred to as a unique “fingerprint”, was identified through the power of AI. Los Alamos team member Christopher Johnson built upon a speech-to-text AI model from Facebook. Instead of translating audio waves into text, his model mapped waveforms analogous to sound waves into an AI model that could predict the incidence of a future stick-slip fault failure. 

Although this experiment was localized to Hawaii, further studies across the globe have also confirmed the growing, powerful presence of AI in seismology. A research team from the Jackson School of Geosciences at UT-Austin produced a promising algorithm which, though not entirely perfect, boasted a 70% accuracy by successfully forecasting fourteen earthquakes. The experiment examined seismologic data in southwestern China over the course of seven months. To formulate predictions, the team relied on an acoustic electromagnetic to AI (AETA) monitoring system, which collected data via electromagnetic and geo-acoustic seismologic sensors. The goal of their experiment? To accurately predict earthquakes an entire week in advance!

Artificial intelligence has carved an optimistic path into the future of geologic sciences and public safety. Not only can earthquake predictions prevent the immediate destruction of property, but they can prepare communities for disastrous subsequent landslides, tsunamis, and other catastrophes. Through AI, global communities can confidently prepare for what once was the most tragic, unpredictable force of nature. The rapidly digitizing world of seismology has the potential to save hundreds of thousands of unsuspecting individuals every year.