(Wired Science)
Tracking Whale Sharks With Astronomical Algorithms
By Brandon Keim May 24, 2011 | 5:25 pm | Categories: Animals, Space, Tech
With the help of algorithms designed to guide the Hubble telescope's starscape surveys, conservation-minded coders have designed software that helps biologists identify whale sharks by their spots. The program enlists the help of citizens with cameras, and lets researchers track Earth's biggest fish across time and oceans.
At the ECOCEAN Whale Shark Photo-Identification Library, people can upload photographs to the database, where they're analyzed and classified. Photographers can then learn about their individual animal and receive emails each time it's spotted. In the meantime, researchers will use the records to study population trends and the histories of individual whale sharks.
"If you put a tag in skin, it wears off or falls away. But we can recognize these animals for the rest of their lives," said whale shark expert Al Dove of the Georgia Aquarium, a participant in the ECOCEAN project. "It lets you recognize and track animals without marking them, and it's permanent."
The program began when Jason Holmberg, then an English teacher in Cairo with a passion for scuba diving, saw his first whale shark during a trip to Djibouti in 2002. Fascinated by the giant, gentle fishes, he accompanied researchers on an expedition that fall.
`If you put a tag in skin, it wears off or falls away. But we can recognize these animals for the rest of their lives.'
To count the animals, they used a technique called mark and recapture: Researchers marked individual whale sharks with plastic tags and, depending on how many were seen in subsequent years, calculated population size and trends. It's a standard technique among biologists, but hard to implement in whale sharks, which tend to shed tags and swim far beyond the scope of any single research group.
"I said, `What percentage of these do you see in the future?' They said, `Less than 1 percent.' And I thought, `There's some room for improvement,'" said Holmberg.
Now a technical writer for EMC, Holmberg wondered whether patterns on whale shark skin, which differ between each individual, could replace the tags. A cold-call enlisted the collaboration of Australian whale shark researcher Brad Norman; Over the next year, they developed a pattern recognition program that mapped X- and Y-coordinates of spots, then compared them between whale shark photographs.
It was a rudimentary approach, but it was a start. Soon the pair were joined by Zaven Arzoumanian, a whale shark enthusiast and NASA astrophysicist who introduced them to the work of Ed Groth, a Princeton University astronomer who'd developed algorithms to compare photographs of the night sky and determine what star patterns they had in common.
A comparison of spot patterns between two whale sharks. ECOCEAN
Groth's equations were developed for astronomers using the Hubble telescope, Holmberg's crew adapted them for biologists studying Earth's biggest fishes. The algorithms did what they hoped to but in a far more elegant way, one that involved calculating the properties of every possible triangle within a pattern's spots, and using those as a basis of comparison. The results were published in a 2005 Journal of Applied Ecology article entitled, "An astronomical pattern-matching algorithm for computer-aided identification of whale sharks Rhincodon typus."
From that work grew the ECOCEAN library, now a database of 32,000 photographs of some 2,800 different whale sharks, contributed by more than 2,600 people — and not just by researchers, but people with cameras who happened to see whale sharks. The largest database of its kind, it's used by whale shark researchers from Mozambique, Belize, Mexico and Australia.
The data is still being calibrated, as it takes years of records before biostatisticians can update equations used with old-fashioned tagging methods and feel confident about extrapolating population trends with the new methodology. But it's very promising, said Dove.
Because whale shark patterns are a permanent tag, they let researchers gather data about single animals over unprecedentedly long times. "If we can see how big an animal was three years ago when we last saw it, then we know the growth rate. That is tremendously useful information, and otherwise hard to learn without some way of perpetually recognizing an individual," Dove said.
"And another thing about whale sharks is that they're a highly migratory species. They don't respect national borders. This opens up the possibility that animals we see can be recognized" anywhere, he added. "If I put a tag on, I only recognize my own, not those of a researcher in a different country. The spot pattern is universal."
The approach isn't limited to whale sharks, either. It's been customized for use on the whisker patterns of polar bears — a species for which standard mark and recapture is especially difficult and possibly damaging — and the fin shape of humpback whales. Later this week, ECOCEAN expects to release a generic, open-source 1.0 version of its of software, which Holmberg hopes will be customized for use by researchers using mark and recapture for any distinctively patterned animal.
With their free software package doing the dirty work of data management, "We want to link computer vision scientists who only want to write algorithms with biologists who are technophobes with statisticians who will use the data to make projections," said Holmberg. "It will be a a framework for studying any species."
Holmberg also hopes that other programmers will follow his lead and lend their coding skills to worthy projects. "Pick the species or concern you're most passionate about, pick the researchers who are working on it, and identify their technical needs," he said. "I'm not even a great programmer. I'm underqualified but highly productive."
0 comments:
Post a Comment