Servizio Comunitario di Informazione in materia di Ricerca e Sviluppo - CORDIS

Synthetic Aperture Radar (SAR) water area

We have developed a method for mapping flood extent from space born Synthetic Aperture Radar (SAR) data. The method is a fully automatic texture based maximum likelihood method. In order to design a near-real time system, we aim at an algorithm that is robust, independent of user input and with low computational demand. The method works well on open surface water. Windy and icy conditions weaken the accuracy. Likewise, the method must be initiated differently in order to correctly classify inundated forests. The method has been implemented in the production line for the FloodMan prototype. The method has been applied to ERS, RADARSAT and ENVISAT ASAR images, covering all FloodMan test areas.

The proposed method is texture based, and uses the local data range, mean and variance as texture features. These texture features are computed from geocoded intensity images, and are logarithmically transformed before utilised in the detector. A simple maximum likelihood (ML) classifier that assumes Gaussian class probabilities is used to discriminate dry land from surface water. The ML classifier is trained from a pre-classification based on thresholding the log-mean image, where the threshold is computed automatically from data. Thus, we do not have to provide training data which accuracy is crucial to the performance of most supervised classifiers. For the supervised surface water detector presented in, training data had to be provided for each scene and incidence angle.

In terms of misclassified pixels, the performance of the proposed method at 23º incidence angle is comparable to the performance of thresholding at 45º, for still open water bodies. Further, the proposed method is robust with respect to changes in topography and slight changes in vegetation types. Thus, we have a well performing, location independent method that is completely independent of user input.

Informazioni correlate

Risultato in breve

Reported by

Norut IT
Postboks 6434
9294 Tromsø