By Sheryl Zhang
When Claire Jasper, a Ph.D. candidate at Columbia University, flashed an X-ray image of a deep-sea sediment core onto the screen, AI took on a whole new meaning beyond self-driving cars or chatbots.
Jasper was talking about rocks and sediment in the cores— what she described as “the trash that the glaciers toss away when they start to retreat and melt.” But the trash reveals the Antarctic ice sheet’s loss of mass. Now, an AI model is helping scientists detect traces of those icebergs buried deep in the sediment clues from the sea floor.
Jasper reported her findings at the annual conference held in southwestern Wisconsin.
During months-long drilling expeditions at Integrated Ocean Discovery Program (IODP) Site U1537 near Antarctica’s Iceberg Alley, Jasper’s team collected hundreds and thousands of meters of sediment cores from the ocean floor. Traditionally, researchers identified iceberg-rafted debris (IRD) by manually counting on X-ray images targeting on very specific intervals, or examining coarser fractions of the debris under a microscope after sieving samples of mud.
IRDs are bits of rock and sediment that icebergs pick up on land and carry out to sea. When the icebergs melt, those particles drop to the seafloor and become part of the sediment record. Because IRD layers mark moments when more icebergs were breaking off and drifting across the ocean, they serve as key clues for reconstructing past climate and ice-sheet behavior.
The surge of AI has given scientists another option.
“You can imagine someone staring at X-ray images and counting grain by grain over hundreds of meters — it’s almost impossible, right? It would take someone a whole lifetime,” Jasper said. “So the advantage of using AI after it’s trained is basically, in two days, I will have a 3-million-year record that spans over 300 meters of core. It’s a huge amount of data that would have taken months or years of very painful work.”
“The solutions are in new technologies,” said Richard Alley, a geologist and professor of geosciences at Pennsylvania State University, commenting on the use of AI in scientific research. “AI may do some really wonderful things for us.”
While Jasper acknowledged AI’s efficiency, she admitted that the tool has blind spots.
She explained that as glaciers move across land, they pick up a wide range of sediment from fine clay particles to large boulders. Her AI model tends to overlook smaller debris in X-ray images since it cannot easily detect grains smaller than about a millimeter.
“I’m not saying that one method is better than the other,” Jasper said. “The AI method is a cool thing to explore and hopefully people will start using it in the future — but there’re a lot of other methods as well.”
Alley, who studies climate and energy systems, noted that while AI tools can accelerate discovery, they also come with significant energy demands and raise environmental, social and ethical challenges.
He said AI is “a little bit energy hungry,” because training large models consumes significant computational power, which leads to electric price shocks.
“You’ve added the demand for electricity without adding the supply, and we’re not rapidly taking advantage of the ones that you can build really fast to supply that demand — and that creates shocks,” Alley said. “Whenever you get a little shortage, you tend to get a large run up in price if there’s no ability to put more on the grid.”