Comparing Object-Oriented Land Cover Classification of Airborne SAR and IKONOS data: A case study of a Belizean Savanna

Abstract

From being relatively under-recognized and under-valued until the past two decades, the neotropical savannas, are now recognised to have a value for their biodiversity in terms of species richness, as natural stores for sequestering carbon and for an expanding range of both timber and forest products. In the light of international agreements such as the Kyoto Protocol, savanna lands mapping, in developing countries like Belize can be directly linked to their economic viability. Remote sensing systems are very effective tools to support management of natural resources, especially in remote or inaccessible areas. In this study the potential of airborne SAR data for interpreting land cover in low density tropical woodlands was evaluated by comparing the information that can be extracted by this data source, to that acquired by more conventional optical data like IKONOS. An object-oriented classification approach was followed using the software eCognition. The two different data sets were segmented and were classified using a nearest neighbour classifier. The classification accuracy was assessed by comparing the results of this automated technique to ground truth data. The results show that both classifications had similar overall accuracies, though the AIRSAR data classification presented higher producer’s accuracies for the majority of the classified land cover classes.