The utility of complex soil reflectance image properties for soil mapping

Al-Hussaini, Abdulrahman (1999) The utility of complex soil reflectance image properties for soil mapping. Doctoral thesis (PhD), Manchester Metropolitan University.

Preview

Available under License Creative Commons Attribution Non-commercial No Derivatives. Download (51MB) | Preview

Abstract

This investigation is concerned with the application of complex quantitative analysis to remotely sensed data for mapping soils. The major aim of this thesis is to examine, by means of illustrative examples, the utility of complex image metrics in the detection, differentiation, and partitioning of satellite images of soil landscapes. Satellite images have been widely used for soil mapping. In order to realise the maximum potential of satellite imagery, improvements are needed both in visual presentation of such images, and in their automatic classification, in order to reveal the complex properties of soil landscape. A Landsat TM image of the Al-Ahsa area of Saudi Arabia was used in the investigation. It presents an ideal region for remote sensing studies due to the absence of vegetation cover and the existence of different type of landforms in a region of low topography. Three techniques for modelling complex elements of images were used and evaluated; Fast Fourier Transform (FFT), Artificial Neural Network Analysis (ANN), Fractal and Multifractal Analysis. The FFT technique developed in this thesis isolates spatial frequency components in specific wavebands. The inverse FFT images are enhanced to (i) display optimised zoning of the image, and (ii) to display specific features. This technique partitions images into major zones that are different zones from the standard soil maps. The ANN technique developed is a non-linear measure of image texture. It shows difference within an image. The texture model is trained on areas selected on the basis of the existing soil map. Substitution analysis of training areas allows an assessment of image zones and boundaries. The texture image is displayed by linear contrast stretch. Zonation does not correspond with published maps or with FFT zonation. The fractal method is based on the local fractal dimension that is used as a texture measure based on a moving pre-set size filter over the entire image. The resulting images do not give zones but shows clear patterns of complexity such as spatial transitions. It is possible to derive areas of similar patterns of transition in complexity. There are implications of these results for soil mapping at the theoretical and practical levels. The implications of the theoretical level are about the existences of soil units defined following the classical approach. In the practical level, the classical approach would be abandoned. There is at present nowhere near the same support of the ideas to complement the traditional mapping approach and raise awareness that soils are inherently complex. The study has important implications for classical theory and practice of soil mapping.