Flood mapping in the Zambezi Region using Synthetic Aperture Radar

Bessinger, Mariel (2016-03)

Thesis (MSc)--Stellenbosch University, 2016.

Thesis

ENGLISH ABSTRACT: Floods occur annually in the Zambezi Region during the rainy seasons, causing economic losses and social disruption. Projected changes in climate and land use could worsen the effects of floods. SAR sensors are active sensors and operate in the microwave region and are therefore not restricted by time of day or inclement weather, making them well-suited for flood monitoring. The aim of this research is to examine the use of ASAR and PALSAR images (with acquisition dates ranging from 17 March 2009 to 30 May 2009) to determine flood extent by classifying open water bodies in the Zambezi Region using two different approaches: binary thresholding and active contour models (ACMs) using the Canny edge detector as initial contour. Classified images were then statistically and visually compared to Landsat images. For ASAR images, overall accuracies ranged between 70% and 99% for the threshold classification method and 58% and 98% for ACMs. For PALSAR images, overall accuracy ranged between 54% and 91% for the threshold classification method and between 60% and 96% for ACMs. Results obtained were adequate for both methods of classification, with thresholds only slightly outperforming ACMs for ASAR images, and ACMs only slightly outperforming thresholds for PALSAR images. These methods are binary classifications, which was acceptable for delineating open water bodies, but flooded vegetation areas were present and methods need to be extended to include these areas. Sensor properties such as wavelength, incidence angle and polarisation have an effect on the effectiveness of identifying flooded water bodies. Longer wavelengths are better suited for open water detection than shorter wavelengths, because of better penetration capabilities than shorter wavelengths. Shallower incidence angles are better suited for open water detection, but steeper incidence angles are better suited for flooded vegetation. HH-polarised imagery performed the best for open water body detection when open water surfaces were smooth, but cross-polarisations performed best when surface roughness was present. Therefore, HH- and HV- polarisations should provide sufficient discriminatory power required for open water and flooded vegetation regions.

AFRIKAANSE OPSOMMING: Daar is jaarlikse vloede in die Zambezi Streek gedurende die reënseisoen wat ekonomiese verliese en ontwrigting in die gemeenskap tot gevolg het. Geprojekteerde veranderinge in klimaat en grondgebruik kan die gevolge van vloede vererger. SAR sensors is aktiewe sensors in die mikrogolf spektrum en word dus nie deur die tyd van die dag of gure weer beïnvloed nie, wat dit baie geskik maak om vloede te kontroleer. Die doelwit van hierdie navorsing is om die gebruik van ASAR en PALSAR beelde, wat tussen 17 Maart 2009 tot 30 Mei 2009 opgeneem is, te ondersoek ten einde die omvang van die vloed vas te stel deur die oopwaterliggame in die Zambezi Streek deur twee benaderings te volg: binêre drumpelwaardes en aktiewe kontoer modelle (AKMs) met die Canny rand detektor as aanvanklike kontoer. Geklassifiseerde beelde is dan statisties en visueel vergelyk met Landsat beelde. Die akkuraatheid vir ASAR beelde het tussen 70% en 99% gewissel vir die drumpel klassifikasie metode en tussen 58% en 98% vir die AKM klassifikasie metode. Die akkuraatheid vir PALSAR beelde het tussen 54% en 91% gewissel vir die drumpel klassifikasie metode en tussen 60% en 96% vir die AKM klassifikasie metode. Die resultate was voldoende vir beide klassifikasie metodes, maar drumpel waardes wat ietwat beter gevaar as AKMs met ASAR beelde en AKMs het ietwat beter gevaar as drumpelwaardes met PALSAR beelde. Hierdie metodes is binêre metodes wat aanvaarbaar was om oopwaterliggame af te baken, maar daar was oorstroomde begroeide gebiede teenwoordig en metodes moet aangepas en uitgebrei word om hierdie gebiede in te sluit. Sensor eienskappe soos golflengte, invalshoek en polarisasie het ʼn invloed op die effektiwiteit van die identifisering van oorstroomde gebiede. Langer golflengtes is meer effektief om oopwaterliggame te identifiseer as kort golflengtes weens die beter deurdringing vermoëns as kort golflengtes. Groter invalshoeke is meer geskik vir oopwaterliggaam identifisering, maar kleiner invalshoeke is meer geskik vir oorstroomde plantegroei. HH-gepolariseerde beelde het die beste gevaar met die identifisering van oopwaterliggame met ʼn gladde oppervlak, maar kruis-polarisering het beter gevaar as die oppervlakte onstuimig was, en daarom sal dit aanbeveel word om beide ʼn HH-gepolariseerde en ʼn HV-gepolariseerde beeld te gebruik om oopwaterliggame te klassifiseer.

Please refer to this item in SUNScholar by using the following persistent URL: http://hdl.handle.net/10019.1/98658
This item appears in the following collections: