The effect of scaling on land cover classification from satellite data

Multisensor data applications have seen considerable development in the field of remote sensing in recent years. The combination of various sources of satellite data opens a new field for research providing different viewpoints on subjects under study. This paper summarises the basic conclusions of work in which the scale of satellite imagery, related to the factor of scale for land cover classification, was investigated. Optical data collected by two different sensors (advanced very high-resolution radiometer and thematic mapper) were tested against the ability to classify correctly specific land cover classes at different scales of observation. It was found that traditional image processing techniques such as resampling can lead to significant differences in the quality of the information extracted. The nature of the spectral signature for some land cover classes may also vary depending on the scale of the. observation and on the type of data used. Understanding the role of scale on the spectral signatures of satellite data will help secure the correct interpretation of any classification results. (C) 2003 Elsevier Science Ltd. All rights reserved.