In our early efforts to develop drone survey methods, we were lucky enough to partner with Insitu Pacific Inc. and test their high-end ScanEagle drone for large-scale surveys of dugongs and humpback whales. This three-year series of trial surveys showed us that aerial imagery surveys were a positive step forward, but highlighted the need for image processing solutions.
Lead researcher
Collaborators
Insitu Pacific
Provided in kind support for the three years of the project, including staff for planning, engineering and operation, as well as the drone (ScanEagle) for trials, including imaging systems, and all support equipment
Natalie Kelly, Australian Antarctic Division, and David Peel, CSIRO Marine Labs
Advised on quantitative approaches and conducting statistical analyses. Provided conceptual advice and developing image processing tools.
Frederic Maire, Queensland University of Technology
Developed a marine fauna detection algorithm to automate the processing of drone images
Eric Kniest, The University of Newcastle
Adapted his mapping software, VADAR, that allowed us to plot images and sightings.

Amanda and the Insitu Pacific Inc. team 2011
Three years of trial surveys
Our trial surveys were conducted in 2010-2012, and our objective was to develop methods for using drones to survey marine mammals. More specifically:
- Trial imaging systems
Trial a suite of imaging systems and determine the most effective and efficient system for detecting and identifying marine mammals, with a focus on dugongs and humpback whales – so firstly, can we actually detect the animals confidently within images of some sort? - Develop standardised method
Given we can detect the animals, what methods do we then employ to ultimately determine distribution and abundance of the animals, i.e., how do we correct for detection probability? - Compare observer surveys to drone imagery surveys
Once we’d come up with the methods, the aim was to directly compare the results from traditional observer surveys and drone surveys of dugongs and humpback whales to test the efficacy of drone survey methods.
Aerial Surveys of Marine Mammals
Researchers commonly survey marine mammal populations to monitor their abundance and distribution. We use this information to determine the important habitat areas of a particular species and ensure their populations are stable.
Aerial surveys provide a broad-scale snap-shot of a population. Aerial surveys have traditionally been used to assess populations over a large spatial scale.
Traditional observer surveys involve flying a small high-wing aircraft at a standard altitude and speed. We fly along parallel transects and we have 4 observers on board the aircraft calling sightings as they see them in real time. A fifth person coordinates the flights and records the sightings.

Why Drones?
Eliminate human risk: Drones totally eliminate the human risk factor involved in having a team of 5 plus the pilot flying low and slow over vast expanses of ocean.
Reduce costs: Aerial surveys are labour intensive and expensive – so we are hoping that drones can reduce costs.
Reduce carbon emissions and noise disturbance: The standard aircraft used for dugong surveys in Australia, the Partenavia P68, consumes approximately 250 times more fuel per hour than the ScanEagle used in these trial surveys. The noise levels of the ScanEagle at 6 metres from the drone are approximately the same as those heard at 9 kilometres from a Partenavia.
Better data: We thought drones would provide more accurate sightings and identification of species, more precise locations and less biased abundance estimates – and there was evidence to suggest we were right.
More flexible: Drones don’t require a regulated airstrip for launch and retrieval and can often be operated from a vessel, providing access to isolated areas never before surveyed. In some cases they may also allow us to survey in a wider range of wind conditions – usually the limiting factor in conducting aerial surveys. See Hodgson et al. 2013.
The ScanEagle
The drone for this project, the ScanEagle, is owned and operated by Insitu Pacific. You can check out the specs for the ScanEagle here.
Trial Series
Trial | Species | Range | Objective |
---|---|---|---|
1 | Dugongs | Within LOS | Test of concept |
2 | Humpback whales | Outside LOS | Detection probability |
3 | Dugongs | 100 km | Direct comparison with observers |
Trial 1
This was our first ever attempt to use drones to survey marine fauna so we just wanted to test whether we could detect dugongs in images at a range of altitudes and whether detection was affected by the environmental conditions. The results of this trial are published in Hodgson et al. 2013.

Trial 2
In the second trial we wanted to determine the probability of humpback whales being detected in the images. So we used the drone to run 17 replicated surveys covering an area that was visible to observers based on a cliff, and we compared the number of whales captured by the drone surveys to the census provided by the observers. We also tested whether we could assess the proportion of time whales spend close enough to the surface of the water to be counted (i.e. their ‘availability’) by filming whale pods from the drone. The results of this trial are published in Hodgson et al 2017.


Trial 3
Following the first 2 trials, the unanswered question was whether the drone would perform as well as humans – i.e. whether the rate of dugong detections in images from the drone is equivalent to detections by human observers under the same conditions. The other key question was, if there are any differences in the rates of detection, are there any particular environmental conditions that explain those differences.
So the objective of this final trial was to fly the drone and piloted aircraft with observers over the same patches of water at the same time so we could directly compare the results. Our results suggest that the drone images afforded more detections, and in particular, larger group sizes, than the observers. Also, the observer group sizes decreased much more dramatically as a result of cloud cover than the group sizes in the imagery. The outcomes of this trial are published in Hodgson et al. 2023.

Image Processing
With our last survey trial, we did 13 hours of flying which resulted in over 60,000 images. Obviously for this method to be viable, there needed to be multiple solutions to processing these images, firstly to pick out potential animals of interest, and secondly to spatially represent the surveyed area and the detections. Luckily we had already forged a collaboration with Frederic Maire and our foray into the world of AI had just begun.
We also developed our early understanding of how drone imagery can be mapped, with the generous input from Eric Kniest who had already written a program called VADAR. He adapted his program so that we could spatially represent the surveyed area and the animal sightings. Basically VADAR allowed us to map all of the images captured using the GPS data and UAV rotation (pitch, tilt and roll) data stored in each image. This understanding led us down the path of developing WISDAM as it is today.