Source: ScienceNetwork Western Australia. Written by Hamish Hastie. Image: Dinesh Valke
REMOTE weed detection is set for major advancement after Curtin University researchers successfully used satellite spectral imagery to pinpoint clusters of one of Australia’s most invasive weed species in the Pilbara.
The research could improve weed management plans for farmers and mining companies working on large areas of land.
The scientists used WorldView-two satellite imagery covering a 430 hectare site, about the size of a small suburb in the Perth metropolitan area, at Mardie Pastoral Station.
They used the imagery to identify populations of the invasive shrub mesquite (a group of several closely related Prosopis species native to the Americas).
The satellite’s eight-band resolution sensors capture visible to near infrared light, which can be used to detect things like vegetation health, leaf mass and moisture content of plants.
Curtin Department of Spatial Sciences researcher Dr Todd Robinson, who led the study, says they successfully differentiated the heavily defoliated mesquite from native vegetation using the satellite’s sensors.
“Each plant has a different spectral signature which means they absorb and reflect electromagnetic radiation differently, but sometimes these differences to can be too subtle to identify with standard remote sensing tools,” he says.
“However with mesquite, because it is defoliated, it’s not reflecting the infrared as much as the coexisting healthy vegetation and this was pronounced in four of the satellite’s eight bands, so that was how we could differentiate it.
“It worked better than we expected.”
Satellite method similar in cost to airborne surveys
Dr Robinson says using new generation satellites to detect light reflection from plants is comparable in cost to standard airborne surveys, but has added versatility.
“Surveys have been mainly airborne until now and the downside to that is generally poorer coverage or additional pre-processing to stitch images together,” he says.
“The satellite can provide much greater coverage and, as it can be tasked to capture imagery very rapidly and you don’t need a pilot.”
Dr Robinson says the technology can be used as a monitoring tool by repeatedly acquiring the same imagery to see how much it changes and how much it is invading the area over time.
He says the next step is expanding the research to a wider area and testing its effectiveness on other plants.
“I think it’s transferrable to many other different species,” he says.
“We would like to look at the entire station and also see if it’s transferrable to another weeds of national significance in the Pilbara, such as Parkinsonia (Parkinsonia aculeate).”
Astron has been working with Worldview-2 and Worldview-3 data for a number of years now. We have successfully delineated Groundwater-Dependent Vegetation (GDV) from non-GDV, and have been able to differentiate Melaleuca from different species of Eucalypt to a high degree of accuracy. We are engaged in an ongoing programme to use coordinated in-situ ground campaigns to collect data used to train an ever-improving machine learning algorithm, which uses both spectral and textural characteristics of vegetation species at multiple spatial scales.
Our current focus is the mixture of visual, infrared (near, short-wave and thermal) and radar data from a range of platforms, including UAV, and satellite, to extract structural as well as spatial and textural characteristics from vegetation communities, to better classify them, and also to quantify their health, density and biomass. For further information on our weed management or geospatial services, please contact our team on 9421 9600.