A time series of 15 Thematic Mapper images was used to analyse the seasonal dynamics of soil spectral signatures in a fluvial corridor. A normalization method based on pseudo-invariant objects was used to compare data. The objective was to identify stable spectral patterns in order to characterize the spectral structure of the landscape. Results revealed that the existing geomorphic units could not be differentiated efficiently by spectral data. Strong spectral confusions persisted through time between most categories. However, several spectral patterns that remained stable through time were observed in the images. Ground investigations revealed that these patterns were characterized by the dominance of one or two soils elements (clay, silt or sand). They helped in the definition of a new typology of the fluvial landscape. A multidate composite image was created from several single images in order to concentrate the information on bare soils (more than 75 per cent of the area was covered). The classification of this image provided a good evaluation of the underlying soil spectral structure of the landscape usually masked by the vegetation cover.
展开▼