Mapping land use/cover in a tropical coastal area using satellite sensor data, GIS and artificial neural networks

A common problem when classifying remotely sensed images in order to map land use/cover is spectral confusion: different land use/cover classes present similar spectral signatures and are misclassified. This paper presents a procedure for mapping land use/ cover combining the spectral information from a recent image and data about spatial distribution of land use/cover types obtained from outdated cartography and ancillary data. Two fuzzy maps, which indicate the membership of each land use/cover class, were generated from the ancillary and spectral data, respectively, using an artificial neural networks approach. The combination of both maps was obtained using fuzzy rules. In comparison with spectral classification, this procedure allowed a statistically significant increase of accuracy of land use/cover classification (from 67% to 79%). The advantages of this procedure for combining spectral and ancillary data, with regard to others previously published in the literature, are that it allows one to take into account previous mapping efforts and to establish relationships between land use/cover and environmental variables specific to the mapped area.