Fall 2024 23 BRETT LOVELACE B I R D S “HOW DO YOU FIND A NEEDLE IN A HAYSTACK, OR IN THIS CASE A MURRELET IN A 2,000-MILE STRIP OF DENSELY FORESTED PACIFIC COAST?” ↑ The unassuming marbled murrelet was a major driver of the 1994 Northwest Forest Plan. vironmentalists during the 1990s “Timber Wars,” the battle over wheth- er to log or preserve the last of the Pacific Northwest’s old-growth trees. Now artificial intelligence developed by Oregon State and the U.S. Forest Service is making it possible to find the bird — and potentially other elusive species such as the Roosevelt elk and red-tailed hawk — more efficiently than ever before. Why does that matter? Until now, there hasn’t been enough information to guide conservation and forest plans. Since finding that a murrelet is nesting in, or occupying, a stand of trees determines whether the trees are protected or can be cut, timber managers often must venture into a forest 20 times over multiple years for pre-logging surveys to confirm that murrelets aren’t present. This new method opens the way to more-informed decisions that could determine whether the murrelet — and the forests on which it depends — survive or disappear. Listening to the Forest How do you find a needle in a haystack, or in this case a murrelet in a 2,000-mile strip of densely forested Pacific coast? Previously, researchers navigated swells on inflatable boats, plucked the birds one by one from the open sea to radio tag them, circled above the forest in roaring planes to pick up their signals and then bushwhacked through thick forests before sunrise to spot their nesting trees. But a team of scientists from OSU and the U.S. Forest Service had a different idea: What you can’t see you might be able to hear. Researchers set up acoustic recording devices to listen to coastal forests. Then they developed a machine learning algorithm to mine the recordings for the murrelet’s most audible call. “Our current work suggests that we have uncovered a survey technique that is not only cheaper, safer for field crews and less invasive to the species, but also increases our ability to detect this iconic bird in forests,” said Adam Duarte of the U.S. Forest Service Pacific Northwest Research Station, lead author of the study. Perhaps surprisingly, the team’s artificial intelligence tool —which is a convolutional neural network, or CNN — analyzes pictures rather than sound. Audio is converted into spectrograms, visualizations that show what pitches are being heard and how loud they are over time. The murrelet’s distinctive “keer” call, when mapped this way, curves like a question mark. The scientists taught the CNN to recognize this shape by showing it verified spectrograms of keer calls. It now identifies murrelet calls correctly more than 90% of the time, making it a highly accurate tool for murrelet monitoring. High-Tech Sleuthing The OSU and U.S. Forest Service team is among the first to develop artificial intelligence tools for monitoring wildlife on a broad scale.The program was originally deployed to monitor spotted owls, and it helped researchers make a population estimate, something that had never been done before. College of Forestry doctoral student Matthew Weldy is now finishing an algorithm that can identify almost 150 additional species. “We’re within striking distance of having the full suite of monitoring services for vocal wildlife species from the Canadian border through the Cascade Mountains and Coast Range of Oregon and Washington to the redwoods of California,” said Damon Lesmeister of the U.S. Forest Service Pacific Northwest Research Station, who led the bioacoustics effort for the study. “It’s hard to get your head around 25 million acres, which is the scale at which we’re working.” Just last year, the team collected 2 million hours of sound from 4,000 acoustic recorders in federally managed forests in the Oregon Coast Range and Washington’s Olympic Peninsula. These data are stored in two supercomputers at OSU’s Center for Quantitative Life Sciences. “If a person were to sit there and work 24/7 listening to those recordings and trying to identify murrelet calls, it would take them 640 years,” said Lesmeister. “It’s totally unfeasible for a human to listen and watch for the birds for that long.” “A lot of people think computers are going to take over biology, but I don’t see it that way,” said co-author Matt Betts, Ruth H. Spaniol Chair of Renewable Resources in the College of Forestry. “In my life, I’ve probably done 6,000 point counts: I have stood in one place for ten minutes 6,000 times to record what birds I hear. These new advancements free me up to do much more interesting and detailed work like catching birds, banding them, getting their survival rates and finding nests. That makes me happy.”
RkJQdWJsaXNoZXIy MTcxMjMwNg==