Assessing land-cover map accuracy for the South African land-cover database

Accurate up-to-date information on land cover and land use is critical for development planning and environmental research. At present, no single, standardised land-cover database exists for South Africa. The CSIR and the Agricultural Research Council (ARC) are jointly producing a single, standardised land-cover database for all of South Africa, Swaziland, and Lesotho.(1-3) The database will be derived using a new data product series of 1:250 000 scale gee-rectified 'SPACEMAP' photographic media images, based on seasonally and climatically standardised (1992-1995) LANDSAT Thematic Mapper (TM) satellite imagery.(4) The SPACEMAPS are currently being constructed to National Cartographic Standards by the Satellite Applications Centre (SAC) of the CSIR to address the demands in map-like instant 'snapshots' of land status for national land-use planning strategies. Digital land-cover data products will be created by manually photo-interpreting the SPACEMAPS using the classification scheme shown in Table 1. The final delta set will be suitable for 1:250 000 scale planning and modelling applications (minimum mapping unit 25 ha), and will be supplied as separate vector GIS coverages equivalent to the standard 1:250 000 map sheet areas as defined by the Surveyor General. To be scientifically credible and fit for use, the database must have an accuracy assessment performed. The purpose of this paper is to discuss the methodology and present a method for assessing small-scale land-cover/land-use map accuracy for the South African National Land-Cover Database, as well as for other small-scale southern African remote sensing projects.