Assessing the utility of Landsat 8 multispectral sensor and the MaxEnt species distribution model to monitor Uromycladium acaciae damage in KwaZulu-Natal, South Africa.
| dc.contributor.advisor | Mutanga, Onisimo. | |
| dc.contributor.advisor | Germishuizen, Ilaria. | |
| dc.contributor.advisor | Peerbhay, Kabir Yunus. | |
| dc.contributor.author | Sheik Oumar, Muhammad. | |
| dc.date.accessioned | 2021-06-25T11:15:05Z | |
| dc.date.available | 2021-06-25T11:15:05Z | |
| dc.date.created | 2020 | |
| dc.date.issued | 2020 | |
| dc.description | Masters Degree. University of KwaZulu-Natal, Pietermaritzburg. | en_US |
| dc.description.abstract | South Africa has approximately 1.27 million hectares of plantation forests, with the forestry industry contributing 1% to the state’s Gross Domestic Product (GDP). A major threat to the industry is an escalating number of tree-damaging insect pests and pathogens. Uromycladium acaciae is a pathogen which causes wattle rust in black wattle (Acacia mearnsii) plantation forests; after its first appearance in 2013 in KwaZulu-Natal, it has since spread to most areas in South Africa where suitable hosts are present, causing severe economic losses to the industry. Traditional field-based methods of assessing forest damage can be labour intensive and time consuming. The effective management of these biotic threats requires quick and efficient methods of assessing forest health. Remote sensing has the potential to assess vast areas of forest plantations in a timely and efficient manner. Therefore, the primary aim of this research is to assess U. acaciae canopy damage using freely available Landsat 8 multispectral satellite imagery and the partial least squares discriminant analysis algorithm (PLS-DA). The study was done on two plantation farms near Richmond, KwaZulu-Natal which are managed by NCT Forestry. The model detected forest canopy damage with an accuracy of 88.24% utilising seven bands and the PLS-DA algorithm. The Variable Importance in Projection (VIP) method was used to optimise the variables to be included in the model by selecting the most influential bands. These were identified as coastal aerosol band (430 nm - 450 nm), red band (640 nm - 670 nm), near infrared (850 nm - 880 nm) and NDVI. The model was run with only the VIP selected bands and an accuracy of 82.35% was produced. The study highlighted the potential of remote sensing to (1) detect canopy damage caused by U. acaciae and (2) provide a monitoring framework for analysing forest health using freely available Landsat 8 imagery. The secondary aim of this study is to use the maximum entropy species distribution model (SDM) to determine potential forestry areas that may be at risk of U. acaciae infection. Species distribution modelling using bioclimatic predictors can define the climatic range associated with the disease caused by this pathogen. The climatic range will help identify high risk areas and forecast potential outbreaks. This study assessed the capacity of the MaxEnt species distribution model (SDM) and bioclimatic variables to estimate forestry areas that have a suitable climate for U. acaciae development. The model was developed using 19 bioclimatic variables sourced from WorldClim. The variables are used as predictors of risk for U. acaciae infection and are applied to the landscape occupied by black wattle plantations. The results produced an area under the curve (AUC) value of = 0.97 suggesting strong discriminatory power of the model. The potential distribution of U. acaciae under future climate conditions was also assessed by applying the model to the bioclimatic variables developed from future climate surfaces acquired from WorldClim. The results emphasized (1) the usefulness of species distribution models for forest management and (2) highlighted how climate change can influence the distribution of U. acaciae due to the expansion and contraction of suitable climatic ranges. Overall, the results from the study indicate (1) Landsat 8 multispectral imagery can be used to detect forest canopy damage caused by U. acaciae, (2) PLS-DA variable importance in the projection can successfully select the subset of multispectral bands that are most important in detecting damage caused by U. acaciae, (3) the MaxEnt species distribution model and bioclimatic variables can be used to identify geographic locations at risk of U. acaciae infection and (4) the variable permutation metric successfully identified the most important bioclimatic variables for U. acaciae development and highlighted the climatic patterns associated with the occurrence of the disease caused by this pathogen. | en_US |
| dc.identifier.uri | https://researchspace.ukzn.ac.za/handle/10413/19507 | |
| dc.language.iso | en | en_US |
| dc.subject.other | Bioclimatic variables. | en_US |
| dc.subject.other | PLS-DA algorithm. | en_US |
| dc.subject.other | Forestry industry. | en_US |
| dc.title | Assessing the utility of Landsat 8 multispectral sensor and the MaxEnt species distribution model to monitor Uromycladium acaciae damage in KwaZulu-Natal, South Africa. | en_US |
| dc.type | Thesis | en_US |