AI and Satellite Technology Revolutionize Invasive Species Detection in Australia

Satellite imagery and artificial intelligence can detect with high accuracy two invasive weed species in Australia, posing a new opportunity for defense against these pervasive plants. Research conducted by Charles Darwin University (CDU) and Charles Sturt University (CSU) explored the potential for SkySat satellite imagery and AI algorithms to detect and map African lovegrass (Eragrostis curvula) and bitou bush (Chrysanthemoides monilifera ssp. rotundata). The study, "Investigating the Potential for the Detection of African Lovegrass and Bitou Bush Using SkySat Earth Observation Satellites," is published in the journal Weed Research. African lovegrass is a highly invasive perennial grass which contributes significantly to the $4 billion per annum required for direct control of all agricultural and environmental weeds. Bitou bush, identified by the Australian Government as a Weed of National Significance, is an aggressive shrub which invades coastal dune vegetation. Bitou bush forms dense thicket to smother native plants and can significantly reduce coastal biodiversity. Detection of these species, particularly African lovegrass, is costly and complicated due to infestations occurring at large scales and in mixed landscapes. The academics fed SkySat satellite imagery of locations across New South Wales into two machine learning algorithms. One model could detect African lovegrass with 89.9% accuracy and bitou bush with 86.1% accuracy. Co-author and CDU Spatial Analyst Glen Shennan, who has expertise in African lovegrass, said this method could become a critical component in the cost-efficient, rapid detection and monitoring of invasive plants in Australia. "Ground sampling is very labor-intensive. If we can make drones and satellites work, it can cut down the cost enormously and you can do this repeatedly to see where the species are spreading," Shennan said. "You can identify vulnerable areas you want to prevent these species from spreading to, and you can direct management funds and mitigation funds to where it's most needed." Shennan said there was a desperate need for rapid, cost-effective methods to detect these species, in particular African lovegrass. "African lovegrass is very opportunistic in the way it grows and can outcompete native grasses," Shennan said. "It's not palatable or nutritious. Sheep and cattle avoid it as far as possible. "There's a lot of work going into managing it, but it is herbicide resistant and the only thing that will kill it, it adapts to very quickly. "It's very fast growing, and grows whenever the weather is right, especially in droughty summers. It likes disturbed ground, so if you have a fire come through, it's the first thing that will come back." Shennan said the accuracy of the results was significant for African lovegrass because it can be difficult to identify. "It looks a lot like poa tussock when it's young. Even experienced botanists have trouble differentiating the two, which is where satellites and some types of drones come in," he said. "They can identify different colors we can't see, and with this we hope we can identify its growth patterns." The study was also authored by CDU Lecturer in Remote Sensing Dr. Richard Crabbe, and CSU Senior Lecturer in Livestock Production Management Dr. Jane Kelly.