More than a million wildebeest migrate each year from the Ngorongoro area in Tanzania across the Serengeti to the Maasai Mara in Kenya and back again, following the rains. Hundreds of thousands of zebra, gazelles and eland mix in with the horned gnus, making this one of the world’s greatest natural wonders.
Counting all those animals, though, isn’t easy. Ground counts are time- and labor-intensive, and the low-flying planes used in aerial surveys can spook the animals, leading to unreliable results as they run for cover. Satellite imagery would seem to be a no-brainer solution, at least at first glance. But using the technology isn’t all that simple. That’s in part because clouds and vegetation easily hide animals from an overhead view. Also, figuring out which dots are animals and which are something else is pretty complicated.
But if you want to try that, the Maasai Mara National Reserve in Kenya is a good place to start: There are sunny skies in the dry season, vast savannahs with few trees and lots of big animals that people would like to know more about.And so that’s where Zheng Yang of the University of Twente in Enschede, the Netherlands, and colleagues focused their pilot study , published December 31 in PLOS ONE . They started with six 50-square-meter patches taken from a high-resolution (0.5 meter) image of part of the reserve taken by the GeoEye-1 satellite on August 11, 2009. Then, working with five experienced wildlife researchers from Kenya, they developed a classification system for the images. The system first identifies pixels that might be animals then uses a technique called object-based image analysis to figure out which pixels are actually animals and which belong to the landscape.
Compared with manual counts, the computer-based method was off by an average of 8.2 percent. In that batch, there were a number of missed animals and objects that were misidentified as animals. And the system was far from perfect. It couldn’t distinguish species, and it might never be able to figure out the difference between similarly sized antelope species, the researchers note. Plus, the team admits that they still need to improve how the system deals with complicated backgrounds and image noise.
But for a trial run of something new that runs in only a few minutes, these results are not bad. Used in combination with traditional counting methods, the technique and its known flaws could prove a useful tool. And if the satellite imagery works for other species in other environments, the researchers note, this could open a “new frontier in ecological monitoring and wildlife conservation.”