By Stephanie Nestor, Assistant Editor, Council Magazine
In an innovative trial, Sunshine Coast Council is testing the capabilities of mapping and AI technology as part of environmental conservation efforts to tackle the spread of invasive plants along the coastline. With the trial still underway, Council is navigating constraints and encountering discoveries, which hopefully promise a new way to understand how to protect native ecosystems.
Although scattered with uniquely beautiful ecological sites along its coastlines, Sunshine Coast Council is routinely dealing with invasive plants that have spread out of control. In an effort to combat the destruction of native flora, Council initiated the Invasive Weeds Project, which aims to trial three mapping technologies to detect invasive plants in the area.
The first stage of the trial involved Aspect UAV capturing images using a drone and the second stage involved a partnership with Queensland University of Technology (QUT) to train artificial intelligence (AI) to recognise the invasive species. Finally, the last stage, still to be completed, will look at how each mapping technique applies to different environmental situations.
The project is funded by the Sunshine Coast Environment Levy in an effort to help Council gain a better understanding of how to manage the region’s biodiversity and find a complementary method to assist identification of invasive plant species.
A garden escapee
During this trial, Council is specifically targeting the broad-leaf pepper (BLP), which is a weed tree that grows in the bushland foreshore areas along the coast between Point Cartwright and Wurtulla.
Originating from South America, the species (Schinus terebinthifolius) was introduced as an ornamental tree, but has since escaped gardens and spread to coastal areas. The tree is considered dangerous to local flora and fauna because it competes with native plants and contains toxic resins that can affect both humans and animals.
Being a relative of the Rhus tree family and poison ivy, the BLP tree sap can cause swelling, itching, rashes, sores and breathing problems, and even eating the leaves and berries can cause gastroenteritis in humans.
BLP is a Category 3 restricted invasive plant under the Queensland Biosecurity Act 2014, which means it cannot be distributed or released into the environment. Yet due to its ability to spread quickly, the tree has grown out of control in the area.
Sunshine Coast Council Invasive Weeds Project Officer, Melissa Hele, said the objective of the project is to explore and trial innovative technologies to assist with identification, surveillance and management of BLP.
“BLP has the potential to impact conservation environments by smothering and transforming ecosystems, outcompeting native plants and reducing the ecological values of natural areas,” Ms Hele said. “Primary spread occurs through seed dispersal by birds and mammals, but they can also reproduce from root suckers and human movement as an ornamental shrub.”
Mapping the coast
The coastal location between Point Cartwright and Wurtulla was chosen for the trial due to challenging on-ground access. During World War II, the area was used for artillery training but the shells did not explode on impact, which means the soil cannot be disturbed just in case one of these ordinances accidentally explodes.
As a result, it is difficult to dig up the area to remove these invasive plants, but with a targeted approach thanks to detailed mapping, it may be easier for Council to manage weeds in the area. Ms Hele said that for each of these mapping types, two plant types will be used as proof of concept – the invasive BLP and the native Pandanus tree.
“Pandanus (Pandanus tectorius) will be used as a ‘simple’ plant type as it is easy to identify and occurs in the same coastal strip. BLP will be the second ‘complex’ plant type, being harder to identify amongst our native vegetation,” Ms Hele said.
“This project aims to create detailed mapping of BLP and Pandanus along an approximately 8km stretch of Sunshine Coast foreshore bushland reserves, including Buddina, Warana, Bokarina and Wurtulla. “We want to create a detailed, high-resolution map of where BLP is in the reserve, and then trial solutions to remove the weed tree.
This is being done in addition to the weed management already occurring along this coastal strip.” The three main mapping types being tested during this trial include:
∞ Orthomosaic imagery (imagery that has been stitched together from a collection of images) and manually classifying these plants within subsets
∞ Identification using AI for both Red Green Blue (RGB), which is standard imagery, and multispectral imagery, which includes additional spectral bands
∞ Ground-truthed sub-sampling to compare the imaging to what is found at the sites
By the end of the trial, Council aims to understand whether AI can successfully identify BLP and Pandanus to a high level of accuracy.
Using the final results from all three mapping techniques, Council can determine the requirements and outputs for each type to better understand the budget, time, and software needed to put this technology into application.
Observations so far
While AI technology is already being used in Australia to identify invasive species, Council are hoping to find new ways to map invasive species to support conservation efforts through high resolution detailed maps that can inform on ground outcomes.
As Council and QUT explore this emerging technology, the use of aerial imagery has already proved useful in surveying the selected trial area for BLP, complementing mapping of the site by ground-truthing.
“Nine sites were selected for subsampling ground mapping, which has been completed by experienced ecological restoration contractors, South East Land Repair,” Ms Hele said. “The points and polygons were mapped for both canopy and sub-canopy for both species. Due to the growth form of BLP, it was easy to identify their trunks to see overall pepper populations.
“Although BLP can be easily identified on the ground by an expert, mapping the broader distribution is challenging. There were difficulties mapping the canopy by ground-truthing due to the growth form and providing access to the sites.
“In many of the foredune areas, native horse-tail she-oaks (Casuarina equisitifolia) obscure the aerial visibility of BLP. Mature pandanus often thrive in conditions of partial shade, so again aerial visibility might be difficult in some areas.”
By combining different mapping types, Council can discover the capabilities and limitations of each technique, which means they can explore these constraints before using the technology on a larger scale. With more information, it may be possible to train the AI to identify more than just BLP and Pandanus.
As of August 2023, all drone imagery has been captured in both RBG and multispectral imagery. QUT have completed labelling orthomosaic imaging for Pandanus, and labelling for BLP is currently being completed. Even though the trial is still underway, Council and QUT are already learning more about the requirements of using this technology.
“The plant identification is being done by QUT using machine learning for both RGB and multispectral. However, the initial percentage accuracy for machine learning identification is not available yet. “Labelling of the orthomosaic is complete for Pandanus, taking about 80 hours, with the lowest orthomosaic resolution for the labelling being 4.4cm.”
For Council, it was important to match the software to one already approved on Council’s systems to make sure the technology was easy to integrate. The workflow has been developed to work within ArcGIS software, which will allow Council to scale models as more data becomes available.
Putting the tech to work
So far during the trial, Council and QUT have gained a better understanding of mapping and data management software applications.
The project will provide for the generation of high-resolution data-sets for training AI detectors. Council hopes this trial on a small section of coastal dunes can be applied to larger areas in the Sunshine Coast region to tackle more invasive plants.
However, many challenges must still be overcome. This includes the correct merging of datasets with different spatial resolutions, and optimising model parameters based on trial and error.
While the Council is still to analyse and release the results, this project demonstrates how experimenting with innovative technology can offer alternative solutions to tackling environmental management. Especially with unique and precarious locations such as the coastline in this trial, part of the experimentation process when using new technology is encountering problems and discovering solutions.