Generation and Assessment of Urban Land Cover Maps Using High-Resolution Multispectral Aerial Cameras

New aerial cameras and new advanced geoprocessingtools improve the generation of urban land covermaps. Elevations can be derived from stereo pairs with highdensity, positional accuracy, and efficiency. The combinationof multispectral high-resolution imagery and high-densityelevations enable a unique method for the automaticgeneration of urban land cover maps. In the present paper,imagery of a new medium-format aerial camera and advancedgeoprocessing software are applied to derive normalizeddigital surface models and vegetation maps. These twointermediate products then become input to a tree structuredclassifier, which automatically derives land cover maps in 2Dor 3D. We investigate the thematic accuracy of the producedland cover map by a class-wise stratified design and provide amethod for deriving necessary sample sizes. Correspondingsurvey adjusted accuracy measures and their associatedconfidence intervals are used to adequately reflect uncertaintyin the assessment based on the chosen sample size. Proof ofconcept for the method is given for an urban area inSwitzerland. Here, the produced land cover map with sixclasses (building, wall and carport, road and parking lot, hedgeand bush, grass) has an overall accuracy of 86% (95%confidence interval: 83-88%) and a kappa coefficient of 0.82(95% confidence interval: 0.78-0.85). The classification ofbuildings is correct with 99% and of road and parking lot with95%. To possibly improve the classification further,classification tree learning based on recursive partitioning isinvestigated. We conclude that the open source software “R”provides all the tools needed for performing statistical prudentclassification and accuracy evaluations of urban land covermaps