Classification of Heather (Calluna Vulgaris) dominated moorland using hyperspectral airborne imagery
Dissertation s0951583.pdf (9.799Mb)
MetadataShow full item record
Considerable research has taken place the last few decades to investigate the utility of remote sensing in monitoring, mapping and classification of vegetation in an overall context. In Northern England and Scotland the most characteristic vegetation type that is found is the Calluna vulgaris (Chapman and Gimingham, 1973; Thompson et al., 1995; Carroll et al., 1999), and even though a range of protocols for surveying has been developed to monitor the condition of Calluna dominated moors, they have been found problematic, including the conventional remote sensing techniques (MacDonald et al., 1998; Egan et al., 2000; Macaulay Institute, 2003; Bonn, 2009; MacArthur and Malthus, 2012). Hyper-spectral images may be the key to the identification of heather moorland age classes since they offer great spectral resolution (Chang, 2003; Fauvel et al., 2013). The objective of this dissertation is to investigate how the accuracy of the classification is being influenced in terms of spatial and spectral scale and if indeed hyper-spectral imagery can be deployed for the classification of Calluna vulgaris age classes. The results suggested that the spectral scale, influences the accuracy of the classification; as the spectral width increases, the accuracy of the classification decreases. The two hyper-spectral datasets successfully identified plant classes and excelled in both classification methods (SVM and SID). Their high spectral resolution results in hundreds of observation channels (Melgani and Bruzzone, 2004; Fauvel et al., 2013), which can be the key for the accurate identification of Calluna age classes. Regarding the spatial scale we concluded that it dramatically affects the accuracy of the classification; even if we use hyper-spectral imagery, the accuracy decreases exponentially as the pixel dimensions are increasing. The ideal case is to have a hyper-spectral dataset with high spatial and spectral resolution in order to accurate classify the Calluna age classes.