Lottering, Romano Trent.Peerbhay, Kabir Yunus.Lubanyana, Andile Njabulo Blessing.2022-10-062022-10-0620212021https://researchspace.ukzn.ac.za/handle/10413/20883Masters Degree. University of KwaZulu-Natal, Pietermaritzburg.Drought is one of the least understood and hazardous natural disasters that leave many parts of the world devastated. To improve understanding and detection of drought onset, remote sensing technology is required to map drought affected areas as it covers large geographical areas. The study aimed to evaluate the utility of a cost-effective Landsat 8 imagery in mapping the spatial extent of drought prone Eucalyptus dunnii plantations. The first objective was to compare the utility of Landsat spectra with a combination of vegetation indices to detect drought affected plantations using the Stochastic gradient boosting algorithm. The test datasets showed that using Landsat 8 spectra only produced an overall accuracy of 74.70% and a kappa value of 0.59. The integration of Landsat 8 spectra with vegetation indices produced an overall accuracy of 83.13% and a kappa of 0.76. The second objective of this study was to do a trend analysis of vegetation health during drought. The normalized difference vegetation index (NDVI) values fluctuated over the years where 2013 had the highest value of 0.68 and 2015 the lowest NDVI of 0.55 and the normalized difference water index (NDWI) had the lowest value in 2015. Most indices showed a similar trend where 2013 had the highest index value and 2015 the lowest. The third objective was to do a trend analysis of rainfall and temperature during drought. The rainfall trend analysis from 2013 to 2017 indicated that the month of February 2017 received the highest rainfall of 154 mm. In addition, July of 2016 received the highest rainfall compared to 2013, 2014, 2015 and 2017 with rainfalls of 6.4 mm, 0.6 mm, 28 mm, and 1 mm, respectively. The temperature trend analysis from 2013 to 2017 indicated that December 2015 had the highest temperature of 28 ° C compared to December of 2013 2014, 2016 and 2017 with temperatures of 24°C, 25°C, 27°C, 24°C, respectively. Furthermore, it was also noted that June 2017 had the highest temperature of 23°C while June 2015 had the lowest at 20°C. The fourth objective of this study was to compare the utility of topographical variables with a combination of Landsat vegetation indices to detect drought affected plantations using the One class support vector machine algorithm. The multiclass support vector machine using Landsat vegetation indices and topographical variables produced an overall accuracy of 73.86% and a kappa value of 0.71 with user’s and producer’s accuracies ranging between 61% to 69% for drought damaged trees, while for healthy trees ranged from 84% to 90%. The one class support vector machine using Landsat vegetation indices and topographical variables produced an overall accuracy of 82.35% and a kappa value of 0.73. The one class support vector machine produced the highest overall accuracy compared to the multiclass SVM and stochastic gradient boosting algorithm. The use of topographical variables further improved the accuracies compared to the combination of Landsat spectra with vegetation indices.enVegetation indices.Remote sensing.Multiclass support vector machine.One class support vector machine.Thesis