Coastal zone landscape classification using remote sensing and model development (Phragmites australis).

摘要

Coastal zone landscape characterization and empirical model development were evaluated using multi-spectral airborne imagery. Collectively, four projects are described that address monitoring and classification issues common to the resource management community. Chapter 1 discusses opportunities for remote sensing. Chapter 2 examines spectral and spatial image resolution requirements, as well as training sample selection methods required for accurate landscape classification. Classification accuracy derived from 25nm imagery with 4m pixel sizes outperformed 70nm imagery with 1m pixel sizes. Eight natural and five cultural landscape features were tested for classification accuracy. Chapter 3 investigated the ability to characterize 1m multispectral imagery into rank-ordered categorical biomass index classes of Phragmites australis. Statistical clustering and sample membership was based upon normalized field-measurements. The red imagery channel showed highly significant correlation with field measurements (p = 0.00) and explained much of its variability (r2 = 0.79). Addition of near-infra red, green, and blue image channels in a forward stepwise regression improved the coefficient of determination (r2 = 0.98). In Chapter 4, a landscape cover map was revised by incorporating expert knowledge into a simple spatial model. Examples are provided for a barrier island environment to illustrate this post-classification methodology. A prototype selection of expert rules was sufficient to change more than 20 per cent of the originally classified landscape pixels. Chapter 5 discusses the development of an empirical model that uses vegetation community classes to estimate: (a) soil type, (b) soil compaction rate, and (c) elevation. Vegetation class proved itself a reliable surrogate for estimating these variables based upon field-based statistical measures of association and significance tests. Vegetation was highly associated with four soil types (Cramer's V = 0.98) and soil compaction rates values at depths of 30 and 46cm (Cramer's V > 0.85), and was able to accurately estimate three decimeter-level elevation zones (r2 = 0.86, p = 0.00). A preliminary model to estimate transverse dune crest heights and locations under forest canopy was presented. Lastly, Chapter 6 offers a summary and concluding statements advocating continued use of remote sensing as an application tool for resource management needs.