What does climate change sound like? With the help of machine learning, scientists have confirmed that it sounds like mate-hungry songbirds in the Alaskan tundra, desperately singing, but missing one another.
According to a research paper published in Science Advances on June 20, machine learning can be used to process over 1200 hours of audio and tell scientists exactly which days and weeks two bird species arrived to the Alaskan tundra to mate. Without the use of machine learning, the researchers claim, the enormous amount data would be impossible to process.
“This approach would enable global-scale understanding of how climate change influences phenomena such as migratory timing of avian species,” the study reads.
The species that they studied—lapland longspurs and white-crowned sparrows—are especially vulnerable to the effects of climate change. They don’t belt out their mating songs unless the weather is relatively mild and snow-free. According to the study, “Breeding songbirds require snow-free patches of tundra to supply critical food and shelter, while cold conditions exacerbate the high energetic costs associated with singing." Lapland longspurs’ calls are described by Allaboutbirds.com as “a variety of zeeps, chips, and rattles. Alarm note a wheezy ‘tee-hu.”; white-crowned sparrrows are “a sharp pink... usually made by males or as an alarm call near the nest.”
The problem is that climate change is warming temperatures in their tundra breeding area, prompting certain birds to migrate earlier in the season. But not all birds migrate earlier, meaning that sometimes, the birds miss each other and don’t have as many chicks.
In order to understand how severely climate change is affecting these birds, the researchers fed six thousand nature sound files to a machine learning system. After the researchers taught the system which parts of these audio clips were bird sounds, the system was able to take that lesson and pick out the bird-mating songs from an astronomically larger data size. Over the course of five bird breeding seasons (May to late June), audio records captured thirty-minute-long recordings spaced several hours apart, four times each day.
By using this method, the researchers were able to confirm that disparities in mating-arrival time is going up, most likely due to climate change. They were also able to confirm that bad weather—from snow cover, to cold temperatures, to severe winds—negatively impact birds’ ability to mate on a day-to-day basis. Without the machine learning system, there was simply no way the researchers could have drawn these conclusions from 1200 hours of audio.
This isn’t the first time machine learning has been applied to decode the natural world. As early as the 1990s, it’s been used to compensate for gaps in biological data, and on the other hand, coherently digest huge amounts of data.
