Abstract
Human-mediated translocation of alien plants began over 12 000 years ago. They were introduced for various purposes, which included ornamental use; however, they end up escaping to areas outside of their original point of entry. Several studies revealed that the majority of alien plant species escaped from cultivation in nurseries and home gardens. These species are able to succeed due to the prevalence of human-aided maintenance, such as the provision of nutrients, water, and pruning; these factors afford them to compete with the native plants. Moreover, alien plant species that succeed and subsequently become invasive have a huge impact on the introduced environment. For example, the 2019 Intergovernmental Platform on Biodiversity and Ecosystem Services (IPBES) report highlighted that alien species accounted for over $400 billion in global economic loss. Moreover, several studies suggested that invasive alien species play a huge role in species extinction and biodiversity decline. While some consider this speculative, the IPBES report indicated that alien species, among other factors, account for approximately 60% of the species extinction globally, with invasive alien species alone accounting for 16% of the
ii
documented animal and plant extinction worldwide. Following this evidence, several studies have documented the impacts of invasive alien species, particularly Invasive Alien Plants, on the environment, economy and livelihood. These impacts cover an array of areas including water, fire regimes, soil nutrients, human health and economic activities such as agriculture and tourism. Additionally, over 10% of South African natural habitats are infested with Invasive Alien Plants, and their impacts are well documented.
Therefore, it is imperative to effectively monitor their distribution and assess their risk at an early stage. To effectively monitor and manage their distribution, it is important to acquire high-resolution data to enable the discrimination of the alien plant, particularly at an early development stage (introduction stage). This need for high-quality data and risk prediction has driven this project, aiming to assess the invasiveness of Equisetum hyemale in South Africa. It is an ornamental plant native to Asia, Europe and America and was introduced in the country through nurseries. To achieve this aim, the following objectives were set: i) To present the spectral library dataset of Equisetum hyemale. ii) To test the utility of machine learning and hyperspectral data to discriminate Equisetum hyemale from morphologically similar species, and lastly, iii) to evaluate the risk the species poses to South Africa. To achieve these objectives, data was collected from nurseries in South Africa, home gardens in the GIS and remote sensing laboratory and through intensive literature search. Data was analysed in R using various statistical approaches. The first step was to generate a spectral library dataset for the alien species, and the results revealed a good quality reflectance of vegetation analysis using hyperspectral data. Secondly, we used the spectral data of Equisetum hyemale along with other three morphologically similar species to test the utility of random forest algorithm to discriminate the alien species from other species. The results revealed a good performance of the model in discriminating species (producer accuracy PA = 87-100%; user accuracy UA = 91-100%; overall accuracy OA = 93%; Kappa coefficient k = 0.91), particularly in 2004 nm in the shortwave region. Finally, the results revealed that Equisetum hyemale is sold in nurseries and planted in home gardens. The species distribution model revealed a good model performance (AUC = 0.938), and the species was predicted to be highly suitable in the eastern and southeastern parts of South Africa. The risk analysis indicated that the species has the likelihood of the species to
iii
survive in the country is probable with moderate consequences. However, due to the fact that the species has been recorded in the country for almost a decade and there was no sign of its naturalised population, it is recommended to apply strict precautions and regulations. The results indicated that the species has the potential to become invasive as it can escape cultivation. Moreover, the availability of the spectral dataset and the ability of machine learning to discriminate between morphologically similar species provide a foundation for early detection and monitoring.