Building a Database of Flood Extension Maps using Satellite Imagery
Roque, Dora1; Afonso, Nuno1; Fonseca, Ana Maria1; Heleno, Sandra2

The lower Tagus River is the Portuguese region which is affected by floods more often, demanding an effort from local authorities to build efficient plans to assure the safety of populations and material belongings. Hydraulic flood models can be useful for this purpose, helping in the identification of risk areas. In order to grant access to accurate models, these must be calibrated using information from past floods. The goal of the present study is the building of a flood extension maps database from the lower Tagus River, occurred between 1992 and 2012, for assisting the model calibration. A set of satellite images from both SAR and optical sensors acquired during flood periods were identified by the team and provided by the European Space Agency (ESA). Although flooded areas are easily identified in optical data, the small number of images available made the SAR information become an important resource. In order to help the SAR image interpretation, field and meteorological data were consulted, together with the comparison between SAR and optical images. A procedure, using object-oriented classification, was developed for the flood delineation in both types of data. Although for optical images a thresholding of the NDVI was enough to provide an acceptable border, for SAR data, three distinct situations had to be considered: calm water surface, water surface affected by turbulence and flooded areas with emerged elements. A radiometric analysis of the images using the logarithm of the backscatter coefficient was used to identify the darkest objects in the images, followed by the study of the classified objects neighbours' altitude provided by a Digital Terrain Model (DTM), which allowed the detection of disturbed water surfaces. The comparison between data from flood and reference times, together with the analysis of the neighbouring areas’ altitude and of the extension of historical floods, provided the identification of non-submerged flooded areas. The usage of the purposed algorithm allowed the identification of 86% of the flooded areas for the considered SAR images.