The interactive effect of spatial resolution and degree of internal variability within land-cover types on classification accuracies

Abstract A study is made to assess the effect of spatial resolution on the degree of internal variability within land-cover classes and then to examine how this within-class variance affects classification accuracy. Airborne Multispectral Scanner data flown at 5 m resolution are degraded to simulate 10 and 20 m data. Classification accuracies within internally homogeneous classes are found to be high at all spatial resolutions. In contrast, classification accuracies of land-cover types characterized by a high degree of internal variability or scene noise improve by up to 20 per cent as spatial resolution is coarsened because the proportion of scene noise is reduced. A further improvement in classification can be achieved by smoothing the imagery prior to classification using various spatial filters. The extent of this improvement was found to be as much as 25 per cent depending on the type of spatial filter used, the window size of the filter, the spatial resolution of the data and the land-cover type bei...