Object-oriented segmentation and classification of wetlands within the Khalong-la-Lithuny a catchment, Lesotho, Africa

Abstract

Wetlands in the Maluti Mountains of Lesotho provide important ecosystem functions which are the basis for valued ecosystem services. However, those functions and values are threatened because of erosion. Assessing the extent of wetlands and their condition is fundamental to developing a strategy for restoration and management, but that's a difficult task given the inaccessibility of the region. Remote sensing is recognized as an effective tool to identify wetland areas and assess erosion. This study adopted object-oriented image segmentation and classification methods to identify wetlands in first-order watersheds forming the headwaters of the Matete River and evaluate ongoing erosion. The object-oriented method achieved an overall accuracy greater than 84%. The results proved that the existing wetland inventory which was built from SPOT 5 images by manual photo-interpretation method overestimated area of wetlands and did not capture the on-going erosion in wetlands. The object-oriented method was useful in determining that gullies are prevalent within the wetlands in each of the sampled watersheds.