Abstract
Invasive Acacia species threaten biodiversity and disrupt the ecological balance of the fynbos biome in South Africa. Their rapid spread alters soil composition, hydrological processes and resource availability impacting native vegetation. Remote sensing has been used to monitor invasive species, with open-source data from satellites like Landsat and Sentinel-2 playing a crucial role. However, there is a need to expand the utility of remote sensing for mapping and monitoring invasive species of specific genera across large spatial extents. This study investigated the utility of multi-sensor remote sensing in mapping and monitoring invasive Acacia species in the City of Cape Town located within the fynbos biome in South Africa.
The first objective of the study was to map Acacia species using Sentinel-2 data and assess the impact of topography on classification accuracy. The results showed that Acacia cyclops was classified more accurately than Acacia mearnsii, as indicated by the producer’s accuracy. The inclusion of topographic variables improved the accuracy of A. mearnsii but reduced the accuracy of A. cyclops. The highest rate of misclassification occurred between the two species, with A. cyclops being more accurately identified at lower elevations, while A. mearnsii was better classified at higher elevations. Variable importance demonstrated the significance of topographic variables (slope, elevation and aspect) in improving A. mearnsii classification accuracy. The second objective was to assess Sentinel-2 data, both alone and integrated with radar, to discriminate Acacia species over 12 months during 2022. Sentinel-2 combined with radar achieved the highest overall classification accuracy for all months. However, Sentinel-2 alone was more effective in distinguishing between A. cyclops and A. mearnsii, while the addition of radar data improved the discrimination of Acacias from other land cover types. Species discrimination was highest during summer and autumn. The third objective was to evaluate the potential of Environmental Mapping and Analysis Program (EnMAP) hyperspectral data in classifying A. cyclops and A. mearnsii. Additionally, the study assessed the impact of radar data and dimensionality reduction. This was achieved using four data scenarios: 1) spectral bands alone, 2) spectral bands combined with radar, 3) EnMAP-derived principal components (PCs), and 4) EnMAP-derived PCs combined with radar. All scenarios achieved high classification accuracies, with spectral bands combined with
ii
radar resulting in the highest performance for species discrimination. In conclusion, this study demonstrated the value of open-source remote sensing data to map and monitor invasive species, with hyperspectral data proving superior for species discrimination. Additionally, phenological variations impact classification accuracy indicating the importance of focusing on specific times of the year for optimal classification outputs. The study also showed the value of integrating optical and radar data to improve classification performance, especially in separating Acacia from other vegetation types. These findings are important for supporting nature conservation efforts and facilitating the development of cost-effective management strategies for threatened ecosystems.