Testing the efficacy of the worldview-2 data in detecting and mapping the spread of fusarium dieback in the urban trees of Johannesburg, South Africa
Biological invasions, such as plant-pests, are a major threat to urban tree species and can have adverse repercussions on urban ecosystem function, detrimental environmental impacts, and further economic implications in metropolitans if left uncontrolled. Understanding the distribution, abundance, and temporality of these pest-related diseases, such as the Fusarium Dieback infestation, are key to assessing the degree of impact on native ecosystems. It is also important that the spread of the disease across large landscapes be tracked and monitored, in order to proliferate and expand management and control efforts. Such undertakings can be achieved using high resolution remotely sensed imagery, such as WorldView 2 (WV-2), and powerful machine learning algorithms such as Random Forest (RF) and Support Vector Machines (SVM), particularly in urbanized areas where there is greater heterogeneity of plant species and therefore higher spectral confusion. This study aimed to test the ability of the eight-band WV-2 multispectral satellite imagery in accurately detecting the occurrences of Fusarium Dieback infestation on London Plane (Platanus × acerifolia) tree species, and subsequently distinguishing these from healthy London Plane (Platanus × acerifolia) trees, and other tree species. The study focused on London Plane tree species in the Northern suburbs of Johannesburg, South Africa, where presence of the Fusarium Dieback infestation is prominent. Additionally, the study also compared the performance of RF and SVM in detecting Fusarium Dieback infestation. The study demonstrates the efficacy of using WV-2 in classifying London Plane trees, with an overall accuracy of 87.40% (0.8363 kappa) and 86.61% (0.8259 kappa) for RF and SVM respectively. Infested London Plane trees were mapped with an overall accuracy of 75% for both RF and SVM, showing the potential of WV-2 in distinguishing Fusarium Dieback at an intra-species level (i.e. occurring within one species of tree). The results of this study can be used to develop predictive models of early detection and warning for larger urban regions in order to aid management efforts and direct national control mechanisms in curbing the spread of the Fusarium Dieback infestation.
A dissertation submitted in fulfilment of the requirements for the degree of Master of Science in Geographical Information Systems and Remote Sensing to the Faculty of Science, School of Geography, Archaeology & Environmental Studies, University of the Witwatersrand, Johannesburg, 2022