Machine learning has revealed a hidden geological landmark beneath south-central Alaska: a razor-sharp, 250-kilometer-long edge of the Yakutat microplate, traced by 1,750 small earthquakes that had never been detected before. Seismologist Meghan Miller of Australian National University and her colleagues deployed temporary seismic stations across the region from 2018 to 2021, feeding their data into a machine-learning workflow that uncovered an earthquake cluster running northwest to southeast in astonishing detail.
The discovery matters because it rewrites our understanding of how the Earth's crust is piecing together—or colliding—in one of the world's most geologically active zones. The Yakutat oceanic plateau sits in what Miller describes as a "tectonic traffic jam," caught between the Pacific plate and the North American plate. This congestion directly shapes the earthquake and volcanic landscape of south-central Alaska, making the precise location and structure of the Yakutat edge crucial to understanding seismic hazards in the region.
What makes this discovery especially striking is how the machine-learning method revealed what traditional seismic analysis had missed. Miller notes that "there's a lot of information hidden in the data that we're now able to extract out that we weren't able to see as easily with more traditional methods." The linear cluster of earthquakes marks where the Yakutat microplate slips directly beneath the North American plate in what geologists call a subduction zone—but without the intervening mantle wedge that typically appears in such collisions.
The newly mapped edge sits directly below the apex of the Alaska Range's curvature and the Denali Fault, the major continental fault system in south-central Alaska. That positioning carries profound implications for earthquake risk. Miller and her colleagues propose that seismic stress from the collision could propagate through the overlying North American plate up to the Denali Fault itself, potentially triggering major earthquakes—including possibly the magnitude 7.9 Denali Fault earthquake that struck in 2002.
The research also reveals compositional differences along the Yakutat plate's extent. West of the "razor edge," tremor signals indicate rock composition that allows slow, continuous slipping where stress cannot accumulate enough to create earthquakes. But the edge itself, marked by the newly detected earthquake cluster, suggests a different composition—one that allows brittle failure and sudden rupture. This distinction helps explain why the same subducting plate behaves so differently in different places.
Perhaps most intriguingly, the newly defined microplate boundaries align with a ring of small volcanic cones around the Yakutat's northern and northeastern margins. That alignment hints that the missing mantle wedge between the Yakutat and North American plates may have started to reestablish itself approximately one million years ago—a process that could reshape the region's geology over the next million years to come.
Miller's research, published in The Seismic Record, represents just the beginning. The next phase will examine earthquakes from before 2018 to extend the record farther back in time, and to map the tectonic configuration farther south toward the Alaskan coast. With machine learning now revealing previously hidden earthquake signals, the full story of the Yakutat's collision with the North American continent is only beginning to emerge.