University of Twente Student Theses


Extracting gradient boundaries using hyper-temporal image analysis: progress towards a tool for gradient analysts

Scarrott, Rory G. (2009) Extracting gradient boundaries using hyper-temporal image analysis: progress towards a tool for gradient analysts.

[img] PDF
Abstract:Environmental gradients overlay each other across the earth’s surface. These environmental heterogeneities are expressed by the plant communities living within them. Some gradients change temporally as well as spatially, reflected in the phenological responses of the local vegetation. Gradient analysts, attempting to understand these gradients, currently have little to aid them to stratify their sampling, along the logic of what they are studying (gradients). Furthermore, current landuse classifications compound the issue, providing little indication of landscape gradients, and giving little aid to the gradient analyst. This exploratory research hoped to produce a piece of logic, capable of guiding research into providing a tool for gradient analysts to stratify their sampling regimes. It investigated whether it was possible to devise a method, to extract gradient boundaries solely from temporal remotely sensed imagery, exploiting hypertemporal NDVI datasets available from the SPOT-VEGETATION instrument. The research consisted of firstly, an appraisal of environmental heterogeneity in the study area, using ISODATA classification, and subsequent NDVI class characterisation with rice crop calendars. Secondly, the research explored two methodologies to extract gradient boundaries from hyper-temporal NDVI datasets. These gradient boundaries were thought to bound regions that were heterogeneous in nature, representing regions over which fluctuations in environmental conditions were eliciting a phenological response in the vegetation. The first methodology was based upon Principal Components Analysis and its ability to summarise the hypertemporal variability. The second, involved a temporal use of Edge Detection, exploiting the asymmetry of time to help define and extract gradient boundaries. Both methods successfully extracted logical gradient boundaries. These did bound regions which were logically heterogeneous in nature, and therefore representative of areas of gradient flux. However, these boundaries, despite coinciding in many areas, did not universally match. Whilst it was evident that both methods are in need of further refinement, it was also evident that the different methodologies extracted boundaries based on different, though entirely gradient-relevant criteria. It was recommended that further research into a hybrid method, should follow research into refining the two attempted methodologies. This does not exclude research into other, alternative methodologies, provided they are founded in gradient logic.
Item Type:Essay (Master)
Faculty:ITC: Faculty of Geo-information Science and Earth Observation
Programme:Geoinformation Science and Earth Observation MSc (75014)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page