Surface water detection using Sentinel radar imagery compared with flood modelling: a case study in Townsville, Australia

Author: Phongeun Thammavongsa

Thammavongsa, Phongeun, 2019 Surface water detection using Sentinel radar imagery compared with flood modelling: a case study in Townsville, Australia, Flinders University, College of Science and Engineering

Terms of Use: This electronic version is (or will be) made publicly available by Flinders University in accordance with its open access policy for student theses. Copyright in this thesis remains with the author. You may use this material for uses permitted under the Copyright Act 1968. If you are the owner of any included third party copyright material and/or you believe that any material has been made available without permission of the copyright owner please contact with the details.


Flooding is a natural disaster that brings about undeniable impacts on the environment and human property. Traditional methods like hydrological and hydraulic models have been developed to provide information regarding flood predictions. The advanced technology of satellite sensors has been used to provide related information to help improve flood predictions, especially in situations of limited ground data. Utilising such technology, this research aimed to explore the usefulness of freely available spaceborne imaging (Sentinel 1 Radar-Radio Detection And Ranging) in providing satellite information that can be compared to flood predictions produced by a hydrological flood model. The research focused on mapping the flood extent that occurred in the urban area of Townsville city, Australia at the end of January to early February 2019, by using both multispectral and Radar images (Sentinel 2 and Sentinel 1 respectively).

The satellite imagery was acquired from Sentinel 1 (C band Synthetic Aperture Radar – SAR) and Sentinel 2 (Multispectral Vis-NIR-MIR wavelengths) which were used for pre-flood (dry state), during and post-flood event (wet states). Pre-processing of the multispectral imagery included geometric and radiometric corrections. Classification of the multispectral images was performed by utilising the Normalised Difference Water Index and a grey scale thresholding. After geometric and radiometric calibration and speckle reduction, the classification of the Radar images was performed using density slicing of the average of VV and VH polarizations, in conjunction with a change detection method.

The overall accuracy of the classification of the two multispectral images, when validated with the map of surface water extent (Landsat Water Observations from Space), was 94.7% and 91.9%, with kappa values of 0.94 and 0.90 (for dry and wet conditions respectively). The overall accuracy of the classification of the Radar image (post-flood event), when validated against the classification of the multispectral image, was 90.0% with a kappa value of 0.87. The qualitative comparison of the classification of flood extent (during flood event) with the map of potential flood depth of Townsville from hydrologic modelling resulted in partial similarities along the river and open water bodies, particularly for areas predicted by the flood modelling to be greater than two metres in depth. The Radar classification showed areas of flooding to the west and north of the area, to which flood modelling was applied. However, differences between the Radar classification and flood modelling were evident in residential areas and these differences are attributed to confusion associated with Radar double bounce from buildings and water, backscatter from objects within the water and from wind-induced rough water surfaces. Longer wavelength SAR, for example, S or L band, could address some of these issues to some extent.

Keywords: Flood, Flood extent, Radar, Sentinel 1, Sentinel 2, Change detection, Threshold, Polarization

Subject: Hydrology thesis

Thesis type: Masters
Completed: 2019
School: College of Science and Engineering
Supervisor: Margaret Shanafield