Landscape from space : geographical aspects on scale, regionalization and change detection

Sammanfattning: Landscapes are complex systems composed of multiscale hierarchically organized entities that interact within unique spatial and temporal scales. These interactions result in scaledependent spatial patterns that visually change depending upon the scale at which they are observed. In order to appropriately monitor, model, and manage our interaction withinlandscapes, landscape analysts require approaches that judiciously integrate complex systems theory, remote sensing data, and image processing capabilities for representation and quantification of landscape characteristics at multiple scales.The principal contribution of this thesis is to propose and develop an approach for the representation and quantification of landscapes at multiple scales. More specifically, a newly created method is proposed that overrides a series of well-documented problems related to regionalization as understood in geography. The main conclusions are that the process of regionalization clearly benefits from explicitly dealing with the scale problem, that fuzzy set theory is applicable for the delineation of regions, that multicriteria-evaluation is a feasible solution for the handling of multiple variables, and that the suggested concept together with high-resolution satellite data is parallel to the manual process in depicting regions.One of the first attempts designed to automatically delineate landscape features at their characteristic scale of expression is reported. This is performed with the use of multiscale object-specific analysis and a novel adaptation of a watershed feature detector resulting in visually meaningful objects. The outcome of this integration is a hierarchical approach that automatically models the emergence of dominant landscape image-objects through scale, from a single scale of remote sensing imagery. This capability is extended and used for multiscale change detection in satellite data.