Collection of endmembers and their separability for spectral unmixing in rangeland applications

Thumbnail Image
Date
2010
Authors
Rolfson, David
University of Lethbridge. Faculty of Arts and Science
Journal Title
Journal ISSN
Volume Title
Publisher
Lethbridge, Alta. : University of Lethbridge, Dept. of Geography, 2010
Abstract
Rangelands are an important resource to Alberta. Due to their size, mapping rangeland features is difficult. However, the use of aerial and satellite data for mapping has increased the area that can be studied at one time. The recent success in applying hyperspectral data to vegetation mapping has shown promise in rangeland classification. However, classification mapping of hyperspectral data requires existing data for input into classification algorithms. The research reported in this thesis focused on acquiring a seasonal inventory of in-situ reflectance spectra of rangeland plant species (endmembers) and comparing them to evaluate their separability as an indicator of their suitability for hyperspectral image classification analysis. The goals of this research also included determining the separability of species endmembers at different times of the growing season. In 2008, reflectance spectra were collected for three shrub species (Artemisia cana, Symphoricarpos occidentalis, and Rosa acicularis), five rangeland grass species native to southern Alberta (Koeleria gracilis, Stipa comata, Bouteloua gracilis, Agropyron smithii, Festuca idahoensis) and one invasive grass species (Agropyron cristatum). A spectral library, built using the SPECCHIO spectral database software, was populated using these spectroradiometric measurements with a focus on vegetation spectra. Average endmembers of plant spectra acquired during the peak of sample greenness were compared using three separability measures – normalized Euclidean distance (NED), correlation separability measure (CSM) and Modified Spectral Angle Mapper (MSAM) – to establish the degree to which the species were separable. Results were normalized to values between 0 and 1 and values above the established thresholds indicate that the species were not separable . The endmembers for Agropyron cristatum, Agropyron smithii, and Rosa acicularis were not separable using CSM (threshold = 0.992) or MSAM (threshold = 0.970). NED (threshold = 0.950) was best able to separate species endmembers. Using reflectance data collected throughout the summer and fall, species endmembers obtained within two-week periods were analyzed using NED to plot their separability. As expected, separability of sample species changed as they progressed through their individual phenological patterns. Spectra collected during different solar zenith angles were compared to see if they affected the separability measures. Sample species endmembers were generally separable using NED during the periods in which they were measured and compared. However, Koeleria gracilis and Festuca idahoensis endmembers were inseparable from June to mid-August when measurements were taken at solar zenith angles between 25° – 30° and 45° – 60°. However, between 30° and 45°, Bouteloua gracilis and Festuca idahoensis endmembers, normally separable during other solar zenith angles, became spectrally similar during the same sampling period. Findings suggest that the choice of separability measures is an important factor when analyzing hyperspectral data. The differences observed in the separability results over time also suggest that the consideration of phenological patterns in planning data acquisition for rangeland classification mapping has a high level of importance.
Description
xii, 93 leaves : ill. (some col.) ; 29 cm
Keywords
Vegetation mapping -- Alberta, Southern , Rangelands -- Alberta, Southern -- Remote sensing , Remote sensing -- Alberta, Southern , Dissertations, Academic
Citation