University of Twente Student Theses


Hypertemporal Vegetation Classification for Flood Hazard Assessment.

Lekoko, Othusitse (2010) Hypertemporal Vegetation Classification for Flood Hazard Assessment.

[img] PDF
Abstract:Floodplain vegetation, its structure and growth properties, affect flow and flood wave propagation during flood events. To estimate the impact of vegetation on the flow resistance, flow conditions are determined for different conditions and properties of vegetation. Traditional methods to determine vegetation properties have mainly been ground based. These methods are proving to be insufficient because of the large spatial heterogeneity of floodplain vegetation. This study looks at remote sensing, specifically at hypertemporal satellite technology, as a promising resource for fast, efficient and more effective method of vegetation characterisation for flood hazard assessment. As a new and emerging field, it is still necessary to collect field data to validate the accuracy of the method, therefore Leaf Area Index (LAI) and landcover characterization were included as part of the study. Nineteen images were used over the Netherlands growing season of seven months. The two week time step between images was very critical not only because of possibility to miss important vegetation growth stages, but also because the floodplain under study has agricultural activities including cattle rearing, haymaking and maize farming which may have noticeable influence on landcover over a short period of time. A questionnaire was distributed amongst the farmers to capture these activities and it was found that most farmers plant their maize in May and harvest in September or October. Grass cutting for haymaking varied on a 4-8 week cycle between farmers, depending on the intensity of the farming activities in a particular farm. In all the cases the grass was left for meadow during the winter period. The floodplain vegetation was classified using three different methods on a 19 layer hypertemporal NDVI stack; Maximum Likelihood Classification (MLC), Spectral Angle Mapper (SAM) and NDVI Based Profile matching. All the classifications performed below expectation with SAM lowest at 50.54% followed by MLC at 54.84%. The classification was benchmarked against the ecotope map which is has the accuracy of 69%. NDVI Profile matching was only assessed visually and no statistical evaluation was performed. The ISODATA output from the NDVI Profile classification showed the subtle differences within landcover that was classified as a unit in MLC and SAM showing potential to take advantage of the temporal dimension within the image stack. Landcover specific NDVI Profiles were produced with the use fieldwork data. The profiles displayed characteristic shape unique to each landcover type and even showed deflection points associated with cropping activities of particular landcover types. The study confirmed that the DMC images can produce landcover specific temporal profiles in the IJssel floodplain but landcover classification using these profiles is not practical. Field measured LAI was related to NDVI without the use transformation coefficients and performed poorly with a low correlation coefficient, R2 of 0.43 for maize, and almost no correlation for herbaceous vegetation for forest. Keywords: Flood Hazard, Hypertemporal, NDVI, DMC, vegetation roughness.
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