Development of a Framework to Assess the Impact of Scale Dependent Factors on the Classification of Landcover Maps

Remote sensing data type, classification technique and class description act together to produce, large differences in the classification of landcover. The resulting map will vary in the extent, patchiness and accuracy of classified areas. Differences in the classification of a landcover map are the result inter-relationships between a number of scale dependent factors such as pixel size, extent and smoothing filters. Many studies have investigated these factors individually using empirical data and have come to conclusions based on their unique case studies without isolating one factor from another. This study holistically investigates the different factors to better understand their interactions and relative importance. The effect of scale dependent factors was tested on presence/absence tree cover maps; a common data layer used in landuse planning worldwide. Extent and pixel size were manipulated and a smoothing filter was applied to examine the differences in classification outcome. The aim of this project was to examine the relationship between scale dependent factors and landscape pattern as measured by total area and landscape metrics. It was found that changes in scale dependent factors affected the level of patchiness however total area remained constant. Furthermore the relationships between the factors generally appeared predictable. The study demonstrated that production of landcover maps can be subjective and that the final product can, to a large degree; result from the classification technique and sensors used.