Comparing semi-automated classification and visual interpretation methods of IKONOS imagery in the context of developing countries - A case study of Belizean Savannas

Abstract

Sustainable natural resources management requires that the geographical distribution and temporal evolution of resources be carefully analysed, and that decisions are proposed and made based on these specific analyses. High quality geographical data is necessary and needs to incorporate several elements including currency, resolution and accuracy. With recent studies identifying significant biomass and hence carbon stocks stored within tropical savannas world-wide, there is a growing need for more accurate mapping of savanna extents and their sub-types, especially when these are dominated by trees).

In this study, the usage of high-resolution satellite imagery (< 1 m/pixel) for monitoring a tropical savanna in Belize, has been evaluated through the analysis of semi-automatic and visual interpretation methods of IKONOS imagery. Computer-based maximum-likelihood classifications were carried out and their accuracy assessed. Thirty visual interpretations performed on an IKONOS subset were then compared to a master interpretation that has been created with local knowledge and advanced image interpretation skills.

The results of the pixel-based MLC were disappointing. With high optical resolution imagery, object-based interpretations (such as visual interpretation) are of higher quality and accuracy than semi-automatic pixel based classifications such as MLC. Despite visual interpretation may be the most appropriate choice as ‘object-based’ interpretation, computerised software like ‘Definiens eCognition’ needs financial investment and training which may not justifiable for the required image analysis operations