Landscape connectivity studies on segmentation based classification and manual interpretation of remote sensing data

An object-based methodology to analyse and quantify connectivity at a landscape level as a general measure is described, and categorical data derived from fused Landsat-ETM imagery and aerial photography is analysed using multiscale image segmentation techniques in eCognition © software.

Abstract

Analysing landscape structure and landscape pattern through indices is relatively widespread in landscape ecology and landscape planning. The lack of comparability of the results between different case studies and across spatial resolutions limits the potential usefulness of landscape metrics, in a context where multi-scale GIS and high resolution remotely sensed data are becoming increasingly available. In this paper, an object-based methodology to analyse and quantify connectivity at a landscape level as a general measure is described. Connectivity is to some degree species-dependent. What is regarded as ‘connected’ for one species can be unreachable for another species. In landscape planning and for many conservation efforts at a landscape level, more general measures of connectivity of patches are needed. Recently some progress has been made in the foundation of fragmentation indices, but connectivity measures are still immature. We analyse categorical data derived from fused Landsat-ETM imagery and aerial photography using multiscale image segmentation techniques in eCognition © software. In an object-based semantic network which links several segmentation levels, rules are formulated to exploit neighbourhood/distance relationships of the resulting patches. The work is done in the context of the European Union funded project BioAssess (EVK2-CT1999-00041).