Mapping sodic patches in the Kruger National Park, South Africa,using object-based image classification techniques on multispectral data

摘要

Sodic patches are defined by the chemical and physical properties of the soil and are nutrient hot-spots that contribute to heterogeneity in the landscape of the Kruger National Park, South Africa. These salt-affected soils produce high quality forage during the wet season and experience intense utilization by herbivores. High sodium soils inhibit plant cover, forming bare soil patches, which can be detected from digital imagery. Landsat 7 ETM+, ASTER and SPOT 5 scenes were classified using object-based image analysis. Spectral differences, topology, geology and a DEM were used to distinguish possible sodic patches from other bare soils, applying different algorithms for each sensor and for images from different seasons. Field validation showed an overall classification accuracy kappa index between 0.61 and 0.78. A map of the location and extent of sodic patches for each management unit for the entire park was produced.