A new local approach for flooding level estimation in urban areas using single SAR images

Abstract

-The aim of this paper is to evaluate the level of flooding in proximity of sensible targets in urban areas using only one Synthetic Aperture Radar (SAR) image. To this purpose a two-step algorithm is here proposed: first the flooded areas are detected in the SAR image; and then the water level is retrieved by inverting scattering models developed for urban areas and now properly adapted for the case at issue. The retrieval is performed through a local approach where the a-priori knowledge of the target ground truth and two gauges in the premises is required. The approach is tested on a High Resolution (HR) TerraSAR-X image acquired during the flooding occurred in the Gloucestershire in July 2007.

Flooding Water Depth Estimation With High-Resolution SAR

“…If the gauges are chosen in the premises of the target, the same building reference height can be assumed (h T A = h T B = h T C ), and therefore, the same local water depth is retrieved (h wA = h wB = h wC ) as considered in [17]. In conclusion, the approach here presented can be regarded as a generalization of the local approach already shown by the authors in [17].…”

“…In particular, the dielectric constant of the wall (ε b ) and that of water (ε w ) to evaluate the scattering parameter S pqw , the angle between the sensor line of flight and the projection of the building wall to the ground ϕ, the radar look angle ϑ, and the height of the local target h before the flooding event and the soil roughness parameters (σ w and L w ) are required to invert the scattering model as already shown in [17]. The uncertainty on the knowledge of these parameters and the adoption of geometric and electromagnetic models are sources of uncertainty for the flood level retrieval; all these factors have been analyzed, and the relative uncertainty has already been computed in [27].…”

“…First of all, a segmentation of the SAR image is performed in order to map the flooded areas and localize the targets of interest (segmentation block). Since the segmentation step is not the primary aim of this paper, we used a simple yet effective method already described in [17], but several alternative procedures can be found in the literature such as those in [34] and [35]. According to Iervolino et al [17], in order to distinguish flooded and nonflooded pixels, low-and high-level thresholds are set according to the statistics of the reflectivity relevant to the flooded areas in the image; in particular, the low threshold T L is set by the mean value intensity μ I diminished by the standard deviation σ I , while the high threshold T H is the sum of the same statistics values…”

“…Within this framework, most of the studies in the literature are about flood detection [13], [15], and near real-time solutions are also suggested [16]. Relying on [17], the information on double scattering between the ground surface and the building wall when the surface is flooded is exploited in [18] to increase the effectiveness at detecting urban flooding in layover regions. Measurements of double scattering radar cross sections (RCSs) are compared to the predictions from an electromagnetic scattering model for the cases of a single image with flooding and a change detection scenario; finally, the results from [18] are included in the flood-map algorithm presented in [16] by the same authors to improve the overall detection performance.…”

“…A different local-based approach for water level estimation based on the availability of just one SAR post-event image was introduced in [17]. The overall complete framework behind that approach plus an additional general method are presented, discussed, and tested in the next sections.…”