Multi Temporal Analysis of Floods and Tsunami Effects: Annotation and Quantitative Analysis

Kurzfassung

This paper addresses the problem of multitemporal analysis of an available TerraSAR-X data time series covering the Sendai region in order to assess flood extent and damages caused by Tohuku-oki tsunami. Over the last decade the use of Earth Observation satellites to support disaster and emergency relief has considerably grown. In order to fully exploit highresolution satellite images, a method based on patches (each image is divided into non-overlapping tiles) is proposed to extract relevant contextual information. The local features of each patch act as a compact content descriptor. Further on, considering the available descriptors, the next step is to cluster the data in order to find similar semantic classes. The SVM classifier implements the concept of query by example using image content. The results include well-defined semantic classes, derived through semiautomatic methods thus developing an effective approach to the multitemporal analysis.