We were back in Shark Bay, WA, in August/September to conduct the ultimate trial of the UAV, and it was a showdown between two teams: the UAV team – six staff from Insitu Pacific, and the Human team – five aerial survey observers and a pilot flying in the Partenavia. Both teams flew the same survey lines (transects) at the same time, so we can finally test whether the UAV images can detect the same number of dugongs that human observers can.
Rest assured both teams took this challenge very seriously. The very experienced team of observers and the pilot love their job, particularly when you get to see places as beautiful and fauna rich as Shark Bay, from the air. But the UAV team were very determined to show the capabilities of their drone and were quick to point out all of its advantages.
My grand plan was to survey all of Shark Bay, using the same survey plan that has been flown during the 5 previous surveys conducted in the Bay since the 1980s (see map below). We had 8 days of flying to achieve this mission, and needed good weather (low winds). To our enormous relief, the weather was kind most of the time.
The most challenging task was for the pilot and the UAV operators to actually start each survey line at the same time. The two were flying at different altitudes and at different speeds (the UAV is a bit slower). The pilot of the Partenavia did an amazing job of accounting for that in timing his holding patterns at the end of each transect. Hats off to both teams – I honestly wasn’t convinced it could be done but through excellent radio communication and planning, they nailed it.
Ultimately I wanted to survey areas where there were a lot of dugongs, because I needed a large sample size in order to compare the sightings of the two teams. The human observers spotted a lot of dugongs in blocks 3, 4 and 5 (see map below), so I although we could have surveyed the remainder of Bay, I decided to survey some of these blocks twice. This means I can investigate the effects of different survey conditions (cloud, wind, glare) on detection rates of dugongs in images versus by humans.
Another member of our field team was Gwénaël, who is developing computer software that will detect the dugongs in the images automatically. This is a vital part of the project – during this field trial we collected 73,000 images, and that is large number of images to manually review! His work is continuing, but he was able to test his algorithms on these new images and advise me on matters such as camera settings and image resolution, so that the images we were collecting would work well with his software.
So who won the showdown? To answer that question I now need to use a combination of manual review and the detection software to find all the dugongs in the thousands of images. This will take quite some time, and I will need to enlist the help of some willing helpers…