Invasive Alien Trees in South Africa's Cape Floristic Region
mapped from Sentinel-2 imagery
Press release
Agricultural Research Council (ARC)
Reading Time: 4 minutes
31 October 2024 – In an ambitious effort to better understand and manage invasive alien plants in South Africa’s ecologically diverse Cape Floristic Region, researchers led by Dr. Alanna J. Rebelo at the Agricultural Research Council have released an Invasive Alien Tree Classification Map for the region. This map, created as part of the NASA BioSCape Project, represents a landmark achievement in environmental mapping using advanced machine learning and freely available satellite technology.
The Threat of Invasive Alien Trees to Biodiversity
The Cape Floristic Region, internationally renowned for its rich biodiversity, faces serious threats from invasive alien tree species. These taxa, such as Eucalyptus, Hakea, Pinus, Populus, Acacia, disrupt natural ecosystems by competing with native plants, reducing water availability, and altering fire regimes. Addressing the spread and impact of these invasive species is crucial for preserving the unique flora and fauna in the region.
“With this project, we hope to provide essential data that can support conservation efforts in a region where the stakes are high,” says Dr. Rebelo. “By combining NASA’s BioSCape Project resources with freely available satellite imagery and machine learning, we’re developing new approaches to identify, track, and support management of invasive alien trees.”
Cutting-Edge Mapping Technology
Using Sentinel-2 satellite imagery, the team employed a Random Forest machine learning classifier to create a 10-meter resolution map identifying six key invasive alien tree classes across the Cape Floristic Region. The classification map provides presence/absence information alongside an uncertainty layer that highlights areas of low classification confidence.
A unique aspect of this project is its reliance on a “pure pixel approach”. This method focuses on Sentinel-2 pixels that represent large, homogenous stands of each invasive taxon, ensuring greater accuracy in detecting areas of significant invasion. Field campaigns held between 2023 and 2024 contributed training data for the model, along with datasets from prior research and community science platforms like iNaturalist. Data collected during the campaigns have also been submitted to iNaturalist, as part of a specially crafted project called “BioSCape Alien Mapping Project”. Citizens can now also make contributions to this project: https://www.inaturalist.org/projects/bioscape-invasive-alien-tree-mapping-project.
Applications and Impact
The BioSCape map has far-reaching implications for conservation, research, and policy development. Resource managers can use the map to target high-invasion areas for restoration efforts, while researchers can leverage it to study IAP impacts on ecosystem services and biodiversity. Furthermore, as part of the NASA BioSCape Project, these data, combined with higher spatial resolution AVIRIS airborne data (5 m) will contribute to a larger body of knowledge on how invasive species affect biodiversity and ecosystem functioning globally.
Accuracy
The team has emphasized that, while the BioSCape map is highly accurate (overall 93%), it should not be considered a definitive inventory of all invasive alien trees in the region. “This is a classification output and not an alien map, and therefore post-processing of this output should be performed in conjunction with the uncertainty data and ideally validated with field visits before making any major conservation or land management decisions,” notes Dr. Rebelo.
Understanding the potential limitations of any classification map is critical and therefore the researchers have included an uncertainty layer with their product. This layer quantifies classification confidence based on bioclimatic and geographic representativeness within the Cape Floristic Region. For instance, classifications in fynbos-dominated areas have higher certainty due to more extensive data collection, while results in less-sampled biomes are flagged for cautious interpretation. The extremes of the region (north-west and east) are also under sampled.
In terms of specific challenges, certain taxa, like Hakea, are overrepresented in most of the region due to their similarity in appearance to other native shrubs, particularly proteas. However, the team took a decision to leave the Hakea results in, as the certain areas are particularly well mapped, particularly in the Hottentots Holland mountain range. The same is true for Poplar, which should be treated with extreme caution as it is heavily over classified. According to technician Nicholas Coertze, however, there are regions, particularly riparian areas within the Klein Karoo, where it is very accurately detected.
The three main focus taxa: pine, gum and wattle are most accurately mapped, however there are cases of potentially serious problems. One example is where wattle is confused with Keurboom (Virgilia oroboides) within the riparian zone at Silvermine Nature Reserve. Another is where there have been recent fires (e.g. in the last few years) where alien invasions cannot yet be detected, for example north of Genadendal and the Caledon Swartberg. Hence the need for ecological knowledge in interpreting this classification output cannot be overstated.
Dataset Accessibility
The research team invites conservationists, ecologists, and government agencies to engage with this resource and use it to inform their ongoing efforts to protect South Africa’s unique natural heritage. It has been made available as a layer on Elsenburg’s Cape Farm Mapper (under “Resource Layers”). The metadata are available on SunScholar: https://doi.org/10.25413/sun.27377211.
Contact
Dr. Alanna Rebelo
Agricultural Research Council, Natural Resources and Engineering, Water Science Unit
Stellenbosch University, Department of Conservation Ecology & Entomology
ASSET Research
Email: rebeloa@arc.agric.za
Website: https://sites.google.com/view/alanna-rebelo
This Post Has One Comment
Very cool. Well done and thanks for sharing!