A new method of vegetation mapping by object-based classification using high resolution satellite data

Abstract

Vegetation, mapping, High resolution, IKONOS, Segmentation, Object, Pixel, Classification Continuous monitoring and updating of distribution maps is essential for conservation and management of plant communities, especially in regions where change and development are rapid. Traditional methods of vegetation mapping rely on interpretation of aerial photographs and follow-up field research, and are expensive, time-consuming and labor-intensive. Remote sensing, using high resolution data such as IKONOS, offers the potential for streamlining the process of producing and updating vegetation maps. In this research, three types of classification; pixel-based maximum likelihood, pixel-based ISODATA and object-based minimum distance, were used to classify IKONOS images obtained for typical countryside landscape in Chiba Prefecture. The results from each of these classifications was compared to a master map based on standards used by the Ministry of the Environment in their physiognomical vegetation maps. The object-based classification outperformed the two pixel-based methods in terms of overall accuracy and Kappa index. In the maximum likelihood classification, over-classification resulted in identification of too many tiny areas; while in the ISODATA method, clustering caused the system to classify several different categories into a single area. The object-based classification, however, produced results that compared well with the master map, indicating that remote sensing can contribute to reduction of the work load and standardization of quality in the early stages of producing and updating vegetation maps.