How, how to keep an eye on wildlife? Inquiring minds thought they knew: observe the beasts and learn to count past two. Then the University of Adelaide came along rubber ducks and a drone.
It's hard to reliably count a crowd. Just ask any protest movement versus the police how many were at a given riot. So how is one to accurately count uncooperative animals?
And who is to do it? Students are snowflakes and citizen scientists get bored after the 19th gnu.
The Adelaide paper, published in the prestigious journal Methods in Ecology and Evolution, explains how lead author Jarrod Hodgson et al tested the theory that counting rubber ducks in aerial photographs taken by a drone is more accurate than "traditional counting methods," meaning wildlife spotters on the ground.
This should not be taken for granted. Software has to be "taught" how to distinguish between a rubber duck and a shoe, for instance.
Why need wildlife be monitored, delight aside? The world is changing fast, humans are the ones changing it and one way as can gauge the changes, and their survivability, is to monitor the state of the wildlife.
To take things to the extreme, if an industrial plant making, say, rubber ducks, pours waste into a local river, and if animals and plants living in or by that river dies, one might suspect that the waste is toxic. But one would only reach that conclusion if one had been watching the wildlife. Setting a trap for instance would only prove that there is/had been one animal, not what the herd is up to, or how the herd is feeling.
Anthropogenic climate change is no longer debatable, it simply is, and poses both threats and opportunities for wildlife. "Knowing how many individuals are in a wildlife population allows informed management decisions to be made," explains the team. This reliable wildlife monitoring is key: extinction rates are already 100-fold greater than the background extinction rate.
It's a bird, it's a drone
Actually, using drones to monitor big animals discernible from the air, such as elephants, seals and flocks of geese, says Hodgson – but nobody knew whether the drone counts were accurate.
To test accuracy, one has to test the technology when one knows the correct answer – how many animals were there to begin with. You know where this is going.
Yes, it is going to making ten rubber duck colonies on the beaches of Port Willunga, Australia using several thousand rubber ducks, which the scientists counted first, obviously. Each colony had a different number of fake ducks.
The ducks (not bath toys but decoys, Hodgson explains) were life-size, for the sake of verisimilitude: 25.5 centimeters by 11.3. "Fortunately for our counters, regular kiddie bathtub toys were too small to replicate Crested Terns which are a medium-sized, common seabird," he told Haaretz. The decoy ducks couldn't fly or waddle about like actual ducks, but did provide some sort of representation of the nesting seabird.
Then, knowing in advance how many rubber duckies there were, the scientists compared the accuracy of counts by spotters on the ground, as opposed to counters using drone imagery.
Saved from deadly boredom by software
Conditions on the day were ideal, writes the team, elaborating that the spotters used binoculars or telescopes to find and count the ducks. Meanwhile, a drone flew over the beach and took photos of the toys from different heights, then citizen scientists were given the photos to count.
"The drone approach won," Hodgson reported. "We found it is more accurate and more precise to have people count birds from the drone imagery than to do it on location."
Wonderful. Drones and people in lab conditions don't get distracted by mosquitoes, for instance.
Thing is, counting animals in a photo is mortally tedious. So the team designed software to count the ducks automatically, which yielded results just as good as humans reviewing the imagery, they write.
Now, it's one thing for software to count dark-complected decoy ducks plunked onto a light-colored sand beach. It's another to count say orangutans swinging in the canopy, or stampeding bisons, let alone distinguish the members of one species from another if they're stampeding together.
It is true, Hodgson acknowledges, that the contrast between the target animal and the background will influence the algorithm's ability to make accurate detections. "However, we have released the code for our algorithm via open access so that other researchers can use and adapt it for their applications," he tells Haaretz. "The algorithm uses training data input by the researcher to determine common features of the target animal relative to the background. This means the algorithm can not only be applied, but also customized, to different animals and habitats."
Further development will hopefully improve the accuracy of the detections, thus further improving the efficiency of this approach to wildlife monitoring, he says. Though there is one snag: animals may be terrified when a buzzing drone flies overhead, which could impair the quality of the census.
Still, this rubber-duck-backed technique could prove crucial. : "Accurate monitoring can detect small changes in animal numbers," Hodgson points out. "That is important because if we had to wait for a big shift in those numbers to notice the decline, it might be too late to conserve a threatened species."