Despite first appearances, Antarctica isn’t all ice and snow. In rocky outcrops and crevices, resilient life—such as mosses and lichens—holds on and even thrives. But mapping where this vegetation lives, and how it’s changing, is a challenge.
SAEF PhD student, Kaelan Lockheart from QUT is working on a way to make this task a lot easier, using drones, artificial intelligence, and a dose of ingenuity.
“The goal of my PhD is to develop a ‘scouting’ unmanned aerial vehicle (UAV) for mapping vegetation, such as lichens and moss, in Antarctica,” Kaelan says.
“By scouting I mean that it will operate autonomously to find and map Antarctic vegetation through real-time deep learning models. To start this work we need to understand how researchers have used UAVs to map Antarctic vegetation in the past as well as draw inspiration from similar fields where real-time mapping has been accomplished.”
Deep learning from past research
His review paper is a deep dive into how researchers have used drones to detect and map vegetation in Antarctica and the Arctic, and an investigation into different models that can be applied to image recognition and classification (also known as semantic segmentation models).
The review found that earlier studies were mainly focused on building the tools and methods; things like drone specifications, sensor designs and data processing techniques. Meanwhile, a recent SAEF study by Kaelan’s colleagues was the first to explore the use of deep learning for analysing data collected by drones in this space.
The big challenge is obtaining labelled data. Deep learning models rely on large datasets of annotated images to learn what to look for, and there is a scarcity of this data from Antarctica, partly because of the challenges in travelling to and conducting fieldwork in the region. To address this, Kaelan explored options like semi-supervised learning models, which could support deep learning models to use unlabelled data.
“Additionally, equipping SAEF’s long-term AIoT monitoring platforms with a multispectral sensor—which strikes the right balance between the detailed hyperspectral sensor and the real-time capabilities of RGB sensors—could facilitate the generation of synthetic data upon which to train the deep learning model,” Kaelan says.
The review also looked at studies which used adaptive path planning, where the drone is fitted with AI which decides where to fly next based on what it sees. This approach is seen as an improvement over static path planning methods (like a lawnmower pattern) because once vegetation was detected, it would enable the drone to fly lower to generate higher-resolution data.
All of this points to the conclusion that all the building blocks for an autonomous scouting drone that can process data in real time are available. “Combining these elements opens exciting opportunities to create a prototype for evaluation in Antarctica, marking a significant step forward in this field,” Kaelan says.
From theory to prototype (via Survivor and Norway)
Kaelan will now develop a model that can detect and classify Antarctic vegetation using existing SAEF datasets and then test its performance in simulated scenarios. From there, he’ll begin integrating adaptive path-planning into the software and build a prototype drone.
“The scouting UAV offers a transformative approach to Antarctic vegetation mapping by improving efficiency and minimizing physical impact on sensitive ecosystems,” Kaelan says.
Eventually, these drones could be integrated with satellites to offer a layered data collection chain and enable precise mapping across large areas, helping to inform conservation policies and monitor and designate Antarctic Specially Protected Areas.
If Kaelan’s name sounds familiar to you, he was runner-up on the most recent season of Australian Survivor. He is currently skiing the length of Norway to raise funds for Cottage by the Sea, an Australian charity that provides holiday opportunities to disadvantaged young people.
Read more
Lockhart, K., Sandino, J., Amarasingam, N., Hann, R., Bollard, B., & Gonzalez, F. (2025) Unmanned Aerial Vehicles for Real-Time Vegetation Monitoring in Antarctica: A Review. Remote Sensing, 17(2), 304. https://doi.org/10.3390/rs17020304