Early last month, on the opening day of bow hunting season here on a swath of swampy state land, a team of researchers slipped bright orange safety vests over their waders so as not to be mistaken for deer. “We’re hunting too,” said Spencer Rhea, an ecologist at Duke University. “Hunting for trees.”

Rhea and his colleagues had gathered to investigate “ghost forests”—otherworldly stands of bleached dead trees drowned by flooding or poisoned by saltwater that is intruding inland. They had been led here by a new map of ghost forests, created with the help of artificial intelligence (AI), that identified individual dead trees standing along the east coast of the United States. Posted as a preprint in May, the map counted millions of dead trees across 36,000 square kilometers of coastal forests, many in areas where ghost forests had not been documented before.

The AI mapping method “does an exceptional job at quantifying just how many coastal forests have already died off,” says Matt Kirwan at the Virginia Institute of Marine Science, who wasn’t involved with the research. “It’s not something that’s happening in the future. It’s happening now,” as climate change raises sea levels, driving flooding from storm surge and pushing salty water further upland.

But the aerial view couldn’t show precisely what killed specific stands of trees. For a closer look, they were braving the muck and mosquitoes of North Carolina’s low-lying Albemarle-Pamlico Peninsula. “It is forensics for tree death,” says Duke ecologist Emily Bernhardt, a leader of the NASA-funded project, which in a nod to Dr. Seuss is called THE LORACS, for Tree Health Evaluated using LiDAR, Optical, and Radar Applications across Coastal Systems. The answers, she says, could help identify other forests at risk of becoming ghosts, a process expected to reduce biodiversity and release planet-warming carbon stored in the trees.

The ghost forest map is largely the result of work by Henry Yeung, a doctoral student at the University of Virginia (UVA). He spent hundreds of hours identifying more than 50,000 dead trees in aerial images of the coastline between South Carolina and Maine. He then used these hand-labeled trees to train a deep learning algorithm to spot their pale color and the distinctive shadows of their bare branches in other images.

Scanning the images on its own, the algorithm counted nearly 12 million dead trees standing near the coast, as well as other patches of ghost forest farther inland along estuaries. That’s likely an undercount, Yeung says, because the method can miss downed trees and dead trees that are hidden by the canopy or thinner than 60 centimeters at their crown.

Continue reading the article on Science.org.

Learn more about the Ghost Forests project.