Terrain classification using Landsat Thematic Mapper and digital topographic data in the Burwash Uplands, Southwest Yukon

Moulton, Joan Elizabeth (1989) Terrain classification using Landsat Thematic Mapper and digital topographic data in the Burwash Uplands, Southwest Yukon. Masters thesis, Memorial University of Newfoundland.

[English] PDF (Migrated (PDF/A Conversion) from original format: (application/pdf)) - Accepted Version Available under License - The author retains copyright ownership and moral rights in this thesis. Neither the thesis nor substantial extracts from it may be printed or otherwise reproduced without the author's permission. Download (20MB) [English] PDF - Accepted Version Available under License - The author retains copyright ownership and moral rights in this thesis. Neither the thesis nor substantial extracts from it may be printed or otherwise reproduced without the author's permission. (Original Version)

Abstract

Landsat Thematic Mapper (TM) images, if analysed properly, can provide land scientists with valuable terrain information. In high relief environments digital classification accuracies to date have been relatively low compared to those in less mountainous terrain. While the nature of the topographic effect on Landsat TM data is not fully understood, it is expected that low accuracy may be attributed, in part, to the lack of an appropriate expression of topography in the Landsat image data set. This study was designed to investigate the influence of various surface cover and topographic parameters on the spectral response measured by the TM sensor and show that a data set composed of topographic terrain descriptors can provide additional information which can be incorporated in terrain analysis of mountainous regions. A second objective was to investigate the improvement in TM terrain classification accuracy that could be achieved for a mountainous area in the Southwest Yukon if an ancillary topographic data set was incorporated in the analysis as a logical channel in a discriminant type classifier. -- Correlation procedures were employed to systematically analyse the relationships between TM spectral response and the topographic component of terrain. Bivariate and multiple correlation coefficients were interpreted to show that landcover and topographic characteristics of the landscape are linked and that the parameters of both these components have an effect on TM data. Canonical correlation coefficients were interpreted to mean that the variance in the sensor data set was not fully explained by the variance in either the surface cover, topographic or combined data sets. This suggested that additional information may be contained in the topographic variables which is not contained in the sensor data and may be useful for classification in high relief terrain. -- Two supervised classification schemes were used to investigate the improvement in terrain classification accuracy that was possible by incorporating topography. These classifications conform to the general principles of the 'landscape approach’ and were based on nine biophysical classes studied in the field and in metric aerial photography. The first classification examined the statistical improvement in classification accuracy that was possible by augmenting spectral TM data with elevation, slope, aspect, relief, and percent vegetation cover measured at 672 pixels in the field. Discriminant functions were generated based on the TM data alone and integrated with the other terrain descriptors in several combinations. Classification accuracy was tested using 102 pixels which had not been used in the derivation of the functions. The results show that overall classification accuracy improved from about 64% when the TM data were used alone to 79% when elevation alone was added and up to 98% when the additional topographic field descriptors were used. Accuracy was 100% when the percent surface cover variables were included. -- The second classification scheme examined the spatial impact of incorporating topography in the classification. This involved a maximum-likelihood classification and mapping of the entire study area using the TM data alone and, subsequently, the spectral plus topographic descriptors extracted from an interpolated digital elevation model (DEM) for all pixels in the study area. Mapping accuracy was 55.8% when the TM data were used alone and 77.6% when topography was incorporated. These results provide evidence that TM and topographic data sets derived from a DEM can be integrated in terrain classification to improve the accuracy of results in high relief environments.

Actions (login required)

View Item