The Department of Conservation is using high-resolution satellite images to detect the pest plant Cuscuta campestris(commonly known as golden dodder) within the Whangamarino wetland and at Lake Whangape.
Whangamarino is an internationally significant Ramsar wetland and faces a range of threats from invasive plants and animals.
DOC Technical Advisor Susan Emmitt says the trial project, undertaken by specialist contractor Fabian Döweler (Bushcraft Analytics), has used machine learning to develop a computer model able to recognise golden dodder within satellite images of the wetland.
High-resolution satellite images provide detail finer than one square metre per pixel, meaning individual objects such as small patches of golden dodder can be seen from space and located precisely on the ground.
All plants and animals reflect light in different ways. The satellite images in this study use part of the light spectrum not visible to the human eye, including near-infrared (NIR) wavelengths, which are particularly useful in identifying golden dodder from other wetland vegetation.
Computers identify and classify the distinctive light reflection pattern of particular plant species (spectral signature) captured in an image.
“This project is an example of remote sensing, allowing crucial conservation data to be gathered from a distance and at a large scale,” Susan says.
“DOC was able to request six specific satellite images of Whangamarino from various vendors to support the work.”
Part of Fabian’s brief was to determine how the use of these specific spectral signatures could identify golden dodder, and the minimum image quality needed.
“Using machine learning models for the analysis means golden dodder can be identified more rapidly and easily in satellite images particularly with the addition of more training data,” Fabian says.
The success of the project suggests the use of the technology could be scaled up to identify golden dodder and other pest plants across wider landscapes, he says.
