Learn how Orbica is using GeoAI to help NZ locate Wilding Conifers and eradicate the pesky pests from our ecosystem.
Orbica helped the Department of Conservation (DOC) to automate the detection of wilding conifers using GeoAI, so they can eradicate them early and protect native New Zealand flora and fauna.
DOC is the government agency charged with conserving New Zealand’s natural and historic heritage. Their vision is for New Zealand to be the greatest living space on Earth by providing a wide range of benefits such as healthy, functioning ecosystems, recreation opportunities and living history.
But there’s a snag: wilding conifers. These pesky pines – an invasive, imported species – disrupt native life by outgrowing and taking up resources from the life around them. They spread quickly and are much more resilient than local tussocks. In fact, DOC estimates that 20 per cent of New Zealand will be overtaken by unwanted wilding conifer forests within 20-years without rapid action.
Our grasslands’ native flora is crucial to protect rare, native birds such as pipit and takahe from predators: when the wilding conifer overgrows a grassland, it diminishes the lower foliage and makes it harder for flightless birds, like the takahe, to escape. So, managing these foreign species is key to keeping New Zealand’s nature in balance.
The spread of wilding conifers is unpredictable, and the problem that DOC faces is being able to identify them as tree saplings, when they’re easy to remove. Traditional methods such as flying helicopters and sending DOC staff out onsite are expensive and slow, especially in alpine areas.
DOC needed a faster and more efficient way to locate the saplings and stop them from spreading.
We suggested trying GeoAI – a combination of deep learning algorithms and geospatial processing that automates feature detection from aerial imagery. DOC supplied us with a sample dataset of tiles from South Island subalpine mountainous areas. We selected the best tiles to build training data from and kicked into gear identifying wilding conifers one by one.
Once we had built up strong indicators for what constituted a wilding conifer on enough tiles to comprise training data, we trained our deep learning feature detection algorithm to recognise them automatically. We found on testing that we could achieve 85% accuracy on the identification and location of wilding conifers in alpine grassland environments – even the tiny wilding conifers that are the most important to weed out.
Now DOC has a successfully proven concept that they can apply to vast areas of alpine grasslands to easily geolocate wilding conifers and root them out before they spread. That gives them more time and resources to keep native New Zealand beautiful!