ABSTRACT
ABSTRACT: In pressured environmentally sensitive and ecologically important areas, such as the Taita Hills study area in Kenya, there is a continuing need for accurate up-to-date and historical land cover mapping to be derived from remotely sensed data, that can be used in change detection studies and for developing sustainable land use policies. However, traditional classification techniques based solely on the spectral response of individual pixels achieve only limited success in complex heterogeneous environments. In an attempt to improve on this situation, multispectral SPOT data from 1987, 1992 and 2003 was subject to an object-oriented classification approach to identify 11 land use/land cover classes derived using the Land Cover Classification System (LCCS) protocol. Ground reference test data was collected to enable accuracy assessment and a comparison with the standard maximum-likelihood technique. The derived maps were used to identify major landscape changes that have occurred in the Taita Hills over the period 1987 to 2003.
