UW News

October 29, 2001

When sperm whales talk, UW researcher listens

When sperm whales talk, Michael Dougherty listens.

Not only that, the University of Washington researcher and electrical engineering doctoral student can recognize the voice and tell you exactly which whale is speaking.

Dougherty has developed a method of acoustic analysis that allows him to identify individual sperm whales by the sounds they emit. The technology, which performed with nearly 100 percent accuracy during a pilot study on a limited number of whales, is believed to be the first of its kind. It promises to help researchers better understand how the leviathans interact as individuals and groups. It could also be an efficient means of tracking the movements of one of the ocean’s more elusive mammals.

“It worked even better than we had hoped during our pilot study,” Dougherty said. “I think biologists who study the whales will find it very useful after we develop it into a field-deployable technology.”

Sperm whales are probably best known to the public through Herman Melville’s novel Moby Dick. Just as the whale in that story was hard to find, so can sperm whales be difficult to keep tabs on in real life. When studying other marine mammals, researchers typically use photography to identify individuals, making distinctions on the basis of such characteristics as marks and coloring or notches and irregularities on dorsal fins or tails.

That method is less effective with sperm whales.

“Whereas other species, such as orcas, will swim together, sperm whales tend to spread out,” Dougherty explained. “They may be a kilometer or more apart. They’re certainly in acoustic contact with one another, but you can’t keep track of who’s where with just a camera.”

To compound the problem, sperm whales tend to be out of sight more often than not. They are considered among the deepest diving of the whales, descending thousands of meters in a single dive and staying submerged for as long as two hours at a time.

“They spend most of their time underwater – about 83 percent of the time,” Dougherty said. “When they come up, it might be a considerable distance from where they went down and it’s hard to get to them in time to take a photo adequate for identification.”
As a result, much remains unknown about how sperm whales behave, he said.

“Most people don’t understand how important it is in studying animals to be able to identify individuals within the group. Acoustic identification could allow us to do that where it really couldn’t be done before.”

For several reasons, an acoustic approach works particularly well in identifying sperm whales, Dougherty said. One is that the whales are extremely noisy.

“They emit short, loud sounds called clicks about once per second,” he said, so researchers have a continuous string of sounds to work with. “It’s very regular – it sounds almost like a clock ticking.”

In addition, the clicks travel well in sea water, allowing scientists to tune in from a distance and listen to more than one whale at a time. Dougherty’s process can differentiate among individuals in a group.

“Acoustic identification can work in all weather conditions, in darkness and at considerable distance during the entire time that the whales spend submerged and out of sight,” Dougherty said. “All vocalizing whales within recording range could be identified simultaneously with a minimum of effort and without disturbing the animals. It offers a lot of pluses.”

Data for the pilot study were gathered during two trips to Norway in 1997 and 1998. Dougherty worked off the country’s northern coast on a whale-watching boat, where the operators made room for him and his equipment during tours. When the boat got close enough to a whale, he would photograph it then record the sounds it made as it dived. That way, he had a record of which whales made what sounds.

Next, Dougherty ran the whale clicks through a computer program that uses an algorithm he developed to analyze features of the sounds in terms of “wavelet coefficients.” Simply put, the program looks at various ways the sound changes during the duration of a click.

“For each whale, we had 100 clicks,” he said. “We used random sets of 50 clicks for training and testing the program. That allowed us to take many testing and training sets from the data to build up statistics on the accuracy.”

With just one click, the computer was able to correctly match a whale about 89 percent of the time. Accuracy quickly improved when multiple clicks were used in the analysis.

With five clicks, the computer scored 98.4 percent for correct matches. That climbed to 99.6 percent for 10 clicks and 99.9 percent for 20 clicks.

“If we listen to enough of a whale’s signal, it becomes virtually flawless,” he said.

Dougherty also refined a side benefit to the process. He can acoustically measure the size of a whale.

The big, blocky head that is the sperm whale’s trademark feature contains a large cavity filled with about 500 gallons of oil. When a click is produced, the sound echoes within that chamber. Researchers can measure the echo time and they know the speed at which sound travels through the oil. That allows them to figure out how large the cavity is. From there, it’s a simple matter to calculate the size of the whale.

“It’s not the main thrust of the research, but it’s a pretty sexy side aspect,” he said.

Dougherty plans to return to Norway soon to gather more data. He wants to test his process on a larger group of animals and some questions still need to be answered, such as how the whales’ signals change during the duration of their dives. He’s also constructing and testing acoustic buoys, which could listen for sperm whales and send data to land-based labs, making monitoring easier and less costly.

“We now know it’s feasible and it’s promising,” Dougherty said. “But we need to find out how robust it is.”

###

Dougherty is in Europe this term collaborating with other scientists on his research, but he can be reached via e-mail at dough@isdl.ee.washington.edu.