Assessment of coastal vegetation degradation using remote sensing in False Bay, South Africa

[摘要] ENGLISH ABSTRACT: The coastal zone, the interface between land and sea, faces much pressure from humanactivities. These coastal pressures make it difficult for the coastal zones to fulfil their naturalfunctions, so threatening the state of coastal environments and making them vulnerable tocoastal disasters and degradation. This study aimed to test whether remote sensing techniquescan be implemented to assess the intactness of terrestrial coastal vegetation at the high spatialresolution required for coastal management. The study focused on the northern False Baycoast, Western Cape, South Africa. The research used is a modification of the methoddeveloped by Lück-Vogel, O'Farrell & Roberts (2013) which involved image segmentationand a habitat intactness index using image derivatives. The procedure used Worldview-2(WV-2) images of high spatial, spectral and temporal resolution acquired on 25 February2014 and 11 October 2014. Both images were pre-processed and segmented into meaningfulobjects using object-based image analysis (OBIA). Five image derivatives and the eightspectral bands were stacked into a single image to extract field-informed training points.Regression analysis was performed on eight spectral bands and five image derivatives toevaluate the most suitable bands to produce a habitat intactness index in a subsequentdecision tree classification. Decision tree classification was generated using two spectralbands, namely the RED and NIR1 bands. These bands were chosen because they gave thebest regression results and they are available in most sensors. The bands were also chosenbecause the study deals with vegetation assessment. The overall accuracy of the results was80.5% which was a satisfactory result with a kappa value of 0.75 (75%) that indicates asubstantial agreement between the remotely sensed result and the reference data. A keyfinding is the importance of seasonality to delineate natural and alien vegetation which isbetter achieved in the dry season. Validation of the results was done using the field-validationpoints of a field visit conducted in June 2016. The output maps generated for habitatintactness consisted of five habitat intactness classes namely highly, moderately and lightlydegraded, intact vegetation and alien vegetation. The output maps can be used to informcoastal managers about conservation at a local scale. It is recommended that validation ofremote sensing results be done in the same season that satellite images were taken.

[发布机构] Stellenbosch University