Applications of object-based image analysis results for the farmland surrounding Kakamega Forest in western Kenya

For 473 km2 of farmland surrounding Kakamega Forest in western Kenya, QuickBird satellite imagery has been analyzed by an object-based image analysis approach. Preprocessing involved atmospheric/orographic correction as well as mosaicing and was followed by ground truthing and visual interpretation. Segmentation was optimized using a newly introduced ‘area fitness rate’ as a discrepancy method and an ‘objective function’ as a goodness method. The final rule set for classification consisted of 831 individual processes and has resulted in the distinction of 15 land use/cover classes. This wealth of information has provided a thorough basis for a) the analysis of land use and landscape structures leading to ten distinct farmland types, and b) the redistribution of census population data via the development of a GIS-based population surface model. The typology and the QuickBird derived houses or the redistributed population have been used to simulate c) alternative futures of rural livelihood considering price development and crop yields, and d) rainwater harvesting potential.