Landscape pattern analysis using spatial autocorrelation measurements of optical remote-sensing data.

摘要

Landscapes are dynamic, complex systems that require a variety of sophisticated observational and analytical techniques for their study. Optical remote-sensing technologies offer a method for large-extent observations of such valuable landscape ecology properties as land-cover type and land feature configuration. Quantitative techniques to analyze the spatial patterns present in these raster-based data have focused on the development of geometric descriptors; however, a statistical approach to the analysis of spatial data offers many benefits to landscape pattern analysis. Of particular interest are measures of spatial autocorrelation. At a global level, these spatial statistical measures characterize the average spatial dependence or heterogeneity characteristics of a landscape. At a local level, they quantify the degree of spatial association at each data site, which, when mapped, identify the spatial distribution of clusters of anomalous values in the landscape. In this thesis, local indicators of spatial autocorrelation are applied to optical remote-sensing data and are examined for their application in three common landscape pattern inquiries: land-cover classification, spatial heterogeneity, and spatial-scale dependencies.; In applying the Getis statistic to a Landsat ETM+ image of Hainan, China, results indicate that extracting the degree of spatial association of spectral values will improve unsupervised class signature development, particularly at smaller neighbourhood sizes and where the global spatial autocorrelation is relatively low. Furthermore, this local measure of spatial autocorrelation provides a method for visually identifying the significance of clusters of high and low spectral values. It therefore provides a statistical technique appropriate for use in augmenting the training stage of a supervised classification.; In applying local measures of spatial autocorrelation (Geary's C i, Getis Gi*, and Moran's Ii) to high spatial resolution, hyperspectral AURORA data of a forested region near Timmins, Ontario, spatio-spectral analysis permits the mapping of categories of spatial homogeneity heterogeneity. This is useful for studies in which the aim is to characterize between- and within-species diversity.; Observational and analytical spatial scales of five different landscape types, as observed by Indian Resource Satellite and Landsat imagery of eastern Ontario, were modified to examine how three common spatial autocorrelation measures autocorrelation (Geary's Ci, Getis Gi*, and Morar's Ii) respond. Results show that as image extent is reduced, only one measure shows a sensitivity. (Abstract shortened by UMI.)