Visualising and tracking uncertainty in thematic classifications of satellite imagery

Satellite sensors impose an artificial gridding on real, complex landscapes as a necessary result of their periodic sampling. This leads to several sources of uncertainty when the digital data are classified, including sub-pixel mixing and possible spatial misregistration. Other uncertainty sources include spectral confusion between distinct landcover types, and bias effects specific to certain sensors. Experts working with remotely-sensed imagery are often familiar and experienced with traditional visualisation techniques. Satellite images are familiar to many users as coloured maps or grey-scale images, and single landcover maps and images can be easily and intuitively explored in many GIS packages. However, there are as yet no standard methods and metaphors for visualising the various sorts of uncertainty in raster satellite imagery, and in products derived from it. This uncertainty can be unevenly distributed in space, and may increase or decrease through propagation as various processing stages are carried out. Users often need to explore their landcover data in order to support practical or policy decisions. visualisation processes, in a format which is understandable and easily queried. Conceptually, fuzzy sets allow the handling of many sorts of uncertainty, as well as the representation of geographic objects which belong, partially or completely, to more than one category. This paper stems from a project (FLIERS) which uses fuzzy classification to model and handle some of the uncertainties mentioned above. This approach creates multi-layered stacks of membership images, which need to be combined and manipulated for easy visualisation. Some promising possibilities for visualising such data have been demonstrated by a number of researchers, and specific tools are implemented in the uncertainty visualisation toolkit described in the present article