Characterising uncertain landscape structure
This theoretically focused thesis investigates the use of soft classification techniques for identifying and quantifying landscape structures in a real landscape (the forest-savannah intergrade in the Beni savannahs, Bolivia). Two soft classification approaches are used in this study: a Semantic Import Model and a Fuzzy Clustering Model. Landscape structures usually are measured using hard classifications and quantified with landscape metrics. However previous research has highlighted problems with the interpretation and reliability of the resulting metric values. Soft classification techniques are more suitable than hard classification techniques for describing landscapes because they can model the internal inconsistencies and vague boundary transitions (ecotones) between patches of landcover that are common in semi-natural landscapes. The `uncertainty' associated with a landcover classification, ignored in hard classifications, was found to contain information about spatial structures such as ecotones. In this work the uncertainty was quantified through the use of a-cuts and landscape metrics. The results show that the soft classifications provide an additional level of information to approaches using hard classifications. The uncertainty was used to map the structure of ecotones which allows: the spatial distribution of ecotones between combinations of landcovers to be visualised; change in landscape structure to be quantified specific to the chosen combination of landcovers. This provides a more detailed and informative analysis of landscape structure and change, especially relevant in semi-natural landscapes characterised by extensive ecotones. In combination with landscape metrics, such mapping techniques improve the ability of landscape ecologists to quantify uncertain landscape structure. Further research is required to establish the ecological relevance of the landscape structures derived from uncertainty information.