Integrating HYMAP airborne hyperspectral data and field-based spectrometer data to map arid zone vegetation

Robinson, K 2008, Integrating HYMAP airborne hyperspectral data and field-based spectrometer data to map arid zone vegetation, Masters by Research, Mathematical and Geospatial Sciences, RMIT University.

Document type: Thesis
Collection: Theses

Attached Files
Name Description MIMEType Size
Robinson.pdf Thesis application/pdf 13.69MB
Title Integrating HYMAP airborne hyperspectral data and field-based spectrometer data to map arid zone vegetation
Author(s) Robinson, K
Year 2008
Abstract During 2002, the Australian Department of Defence hosted a high explosives trial at the Woomera Large Scale Explosives Test Area (LSETA) in South Australia. HyMap® airborne hyperspectral imagery and field-based spectrometer data were simultaneously collected across the trial site. The data was used to test the capability of the hyperspectral sensor to map the sparse Chenopod Shrubland using three common classification algorithms, Spectral Angle Mapper (SAM), Spectral Feature Fitting (SFF) and Mixture Tuned Matched Filter (MTMF).

The HyMap® dataset was atmospherically calibrated by two methods: the Empirical Line Method; and a radiative transfer method, ATREM, cascaded with the Empirical Line Method. The calibrated imagery was classified by reference to a spectral library containing field-derived spectra of nine local materials. Using ground-truth data points, collected during the field campaign, the classification result was poor. Of the three algorithms, the SAM attained the best overall percentage accuracy, the best producer and user percentage accuracies, and the best Kappa coefficients.
When the classification maps were compared with high-resolution, pan-sharpened, multispectral Quickbird® satellite imagery, the assessment of the classification maps indicated that the results were better than the statistics indicated with plant species associated with the expected geomorphologic features. Canegrass (Eragrostis australasica) was generally located in the swamps and gilgai; two of the three saltbush species represented, Atriplex vesicaria and A.macropterocarpa, were located on the plain, while the third, A.lindleyi, was located within a swamp; and the Spikeycottonballs (Dissocarpus paradoxa) was spread throughout the landscape. The soils and Gibbers appeared to coincide with the appropriate landcover classes within the landscape.

Three primary tasks were undertaken to produce the classification maps: imagery calibration, imagery classification and accuracy assessment.

At least six factors were identified as contributing to the outcome of the classification, including co-registration of the ground-truth data with the imagery, inappropriate sampling of the ground-truth data, and unseasonal weather conditions during the fieldwork. There is little doubt that the sparseness of the vegetation contributed to the disappointing result; however, it is likely that the response of the vegetation to environmental factors also contributed. During the calibration of the imagery, the ATREM/ELC cascade was used to determine if the inclusion of a radiative transfer model would better accommodate the bi-directional reflectance (BDR). Differences between the perpendicular flight-line classification maps suggest that the accommodation was not complete.

The outcome indicates that for a true assessment of the utility of hyperspectral imagery in mapping arid zone vegetation, several conditions need to be met: (1) the geo-location of the imagery and ground-truth data needs to be ±0.2m or better due to the small tolerance for error. The use of high-accuracy GPS methods is indicated; (2) imagery calibration needs to take into account the BDR of the whole scene, including the vegetation; (3) the standard of data used to assess the quality of the classification results needs to be high. (4) appropriate statistical methods need to be developed; (5) environmental responses of the vegetation to the arid conditions need to be taken into account.
Degree Masters by Research
Institution RMIT University
School, Department or Centre Mathematical and Geospatial Sciences
Keyword(s) Australia
South Australia
vegetation classification
arid zones
Version Filter Type
Access Statistics: 360 Abstract Views, 1259 File Downloads  -  Detailed Statistics
Created: Thu, 09 Dec 2010, 15:33:56 EST
© 2014 RMIT Research Repository • Powered by Fez SoftwareContact us