Algorithm Raises New Questions About Cascadia Earthquake Records

The Cascadia subduction zone in the Pacific Northwest has a history of producing powerful, destructive earthquakes that have engulfed forests and generated tsunamis that have reached as far as the coast of Japan.

The last major earthquake occurred in 1700. But it probably won’t be the last. And the areas at risk of being hit are now bustling metropolises that are home to millions of people.

Determining how often earthquakes occur – and when the next “big” one will occur – is an active scientific question that involves looking for signs of past earthquakes in the geological record in the form of rocks, sediments and shaken landscapes.

However, a study by scientists at the University of Texas at Austin and their collaborators calls into question the reliability of a record of earthquakes that spans thousands of years – a type of geological deposit called turbidite that is found in layers of the seafloor.

The researchers analyzed a selection of turbidite layers from the Cascadia subduction zone dating back about 12,000 years with an algorithm that assessed the correlation between turbidite layers.

The researchers found that in most cases, the correlation between turbidite samples was no better than random. Because turbidites can be caused by a range of phenomena, not just earthquakes, the results suggest that the link between turbidite records and past earthquakes is more uncertain than previously thought.

“We want anyone who cites the time intervals of Cascadia subduction earthquakes to understand that those chronologies are called into question by this study,” said Joan Gomberg, a research geophysicist at the U.S. Geological Survey and co-author of the study. “It’s important to do more research to refine those intervals. What we do know is that Cascadia has been seismically active in the past and will be in the future, so ultimately, people need to be prepared.”

The findings do not necessarily change the estimated frequency of Cascadia earthquakes, which occur about every 500 years, the researchers said. The current frequency estimate is based on a range of data and interpretations, not just the turbidites analyzed in this study. However, the findings highlight the need for more research on turbidite layers, in particular, and how they relate to each other and to large earthquakes.

Jacob Covault, co-author and research professor in the School of Geosciences at the University of Texas at Jackson, said the algorithm offers a quantitative tool that provides a reproducible method for interpreting records of ancient earthquakes, which are typically based on more qualitative descriptions of the geology and their potential associations.

“This tool provides a reproducible result, so everyone can see the same thing,” said Covault, co-director of the Quantitative Clastics Lab at the Jackson School’s Bureau of Economic Geology. “There’s potential for disputes about that result, but at least you have a baseline, a reproducible approach.”

The results were published in the journal Bulletin of the Geological Society of AmericaThe study involved researchers from the USGS, Stanford University and the Alaska Division of Geological and Geophysical Surveys.

Turbidites are the remains of submarine landslides. They consist of sediments that have been deposited on the seabed after being thrown into the water by the turbulent movement of sediments rushing along the ocean floor. The sediments in these layers have a particular gradation, with coarser grains at the bottom and finer grains at the top.

But there are several ways that a turbidite layer can form. Earthquakes can cause landslides when they shake the seafloor. But storms, floods, and a whole range of other natural phenomena can also cause them, although on a smaller geographical scale.

Currently, to link turbidites to past earthquakes, they typically have to be found in geological cores taken from the sea floor. If a turbidite appears in roughly the same location in multiple samples over a relatively large area, it is considered a remnant of a past earthquake, the researchers say.

Although carbon dating the samples can help clarify the chronology, there is still much uncertainty in interpreting whether samples that appear at roughly the same time and place are related by the same event.

A better understanding of the relationship between different turbidite samples prompted researchers to apply a more quantitative method to the turbidite data, an algorithm called “dynamic time warping.” This algorithmic method dates back to the 1970s and has a wide range of applications, from speech recognition to smoothing graphics in dynamic virtual reality environments.

This is the first time it has been applied to turbidite analysis, said co-author Zoltán Sylvester, a research professor at the Jackson School and co-principal investigator in the Quantitative Clastics Lab, who led the adaptation of the algorithm for turbidite analysis.

“This algorithm has been a key part of many projects I’ve worked on,” Sylvester said. “But it’s still very little used in the geosciences.”

The algorithm detects the similarity between two samples that may vary over time and determines how well the data between them matches.

For speech recognition software, this means recognizing key words even if they are spoken at different speeds or pitches. For turbidites, this means recognizing shared magnetic properties between different turbidite samples that may appear different from one location to another even though they come from the same event.

“Correlating turbidites is not a simple task,” said Nora Nieminski, co-author of the report and coastal hazards program manager for the Alaska Division of Geological and Geophysical Surveys. “Turbidites typically exhibit significant lateral variability that reflects their variable flow dynamics. Therefore, turbidites are not expected to maintain the same depositional character over large distances, or even small distances in many cases, particularly along active margins like Cascadia or in diverse depositional environments.”

The researchers also subjected the correlations produced by the algorithm to another level of scrutiny. They compared the results to correlation data calculated using synthetic data obtained by comparing 10,000 pairs of random turbidite layers. This synthetic comparison served as a check against chance matches in the real samples.

The researchers applied their technique to magnetic susceptibility logs of turbidite layers in nine geological cores collected during a scientific campaign in 1999. They found that in most cases, the connection between turbidite layers that had been correlated before was no better than random. The only exception to this trend was for turbidite layers that were relatively close to each other, only 24 km apart.

The researchers stress that the algorithm is just one way to analyze turbidity, and that including other data could change the degree of correlation between cores one way or another. But according to these results, the presence of turbidity at the same time and in a general area of ​​the geologic record is not enough to definitively link them together.

And while algorithms and machine learning approaches can help with this task, it is up to geoscientists to interpret the results and see where the research leads.

“We’re here to answer questions, not just apply the tool,” Sylvester said. “But at the same time, if you’re doing this kind of work, it forces you to think very carefully.”