You really can count on drones
Can you trust a drone? For some time now, we've used UAVs to assist with the monitoring of animal populations for conservation efforts, but we just kind of assumed they were up to the task. The good news is that our hunch appears to have been correct, according to a new paper from the University of Adelaide.
"With so many animals across the world facing extinction, our need for accurate wildlife data has never been greater,” said the study's lead author Jarrod Hodgson. “Accurate monitoring can detect small changes in animal numbers. 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.”
To test drones' effectiveness, you can't use real wild animal colonies, because, by its nature, the number is uncertain. So instead, Hodgson's team created ten replica seabird colonies comprising of several thousand birds. Rubber ducks (not the bright yellow bathtub variety) were dotted around an Adelaide beach to replicate the colonies of the Greater Crested Tern.
Colonies ranged from 463 to 1,017 rubber ducks. That's a lot of ducks, as you can see in the video below.
All ducks in a row, it was down to the test: would experienced human wildlife spotters be best with their telescopes or binoculars, or would drone photography trump them? While the spotters did things the old-fashioned way, citizen scientists tallied up the figures from drone photography, and it was the latter that proved more accurate.
More accurate, yes, but you're still looking at a considerable human element involved. Whether on the bank opposite the colonies, or examining photography remotely, squishy, fallible humans are still required. Could there be a way of doing cutting out the middleman? Yes. With a little algorithmic training, the researchers were able to train a programme to automatically extract results every bit as accurate the human count.
“We used increasing amounts of training data to train independent algorithms,” Hodgson tells me via email. ”This allowed us to investigate the relationship between the amount of training data required, and therefore user time, and accuracy. We identified 10% training data as a threshold above which little improvement in count accuracy was achieved for our application.”
The code for the algorithm is available via open access, should other researchers with to modify it for their own projects. “The algorithm can be not only be applied but also customised to different animals and habitats,” Hodgson explains. “We expect that with continued development, the accuracy of the detections can be improved which will further improve the efficiency of this approach to wildlife monitoring.”
There is another advantage to drone-based photography over human wildlife spotters: unlike the rubber ducks on the Adelaide beach, real birds move around. “In the field, ground counters contend with the movement of live birds while counters of drone-derived imagery use static images,” Hodgson explains. “That source of error is minor or non-existent for counts made from drone-derived images.”
Regardless, there's a long list of birds and mammals that can benefit from the time-saving accuracy of this new system. Everything from pelicans to dugongs can be tracked more easily, Hodgson reckons, and “there has also been promising results using the nests or tracks of animals, such as orangutans and turtles, to infer presence.
“In time, with further research and development, it is likely that drones will carry a variety of powerful sensors that will give ecologists and environmental managers an unprecedented insight into the animal world.”