Assessment of a SIFT-based DEM extraction approach using GEOEYE-1 satellite stereo-pairs in flood mapping of an ungauged basin

A module for Digital Elevation Model (DEM) extraction from Very High Resolution (VHR) satellite stereo-pair imagery is developed and used for a flood plain mapping applications at an ungauged basin. A procedure for parallel processing of cascading image tiles is used for handling the large datasets requirements of VHR satellite imagery. The Scale-Invariant Feature Transform (SIFT) algorithm is used to detect potentially homogeneous features in the members of the stereo-pair. The resulting feature pairs are filtered using the RANdom SAmple Consensus (RANSAC) algorithm by using a variable distance threshold. Finally, homogeneous pairs are converted to point cloud ground coordinates for DEM generation. A 0.5 m × 0.5 m Geoeye-1 stereo-pair acquired over an area of 25 km2 in the island of Crete, Greece is used as input for the module. The resulting 1.5 m × 1.5 m DEM has superior detail over previously developed 2 m and 5 m DEMs that are used as reference, and yields a Root Mean Square Error (RMSE) of about 1 m compared to ground truth measurements. 1D and 2D hydraulic models are used to simulate the rainfall-runoff characteristics and the flood wave kinematics, of the flash flood event of October 17th, 2006 that occurred in Almirida basin, by using the 1.5 m VHR-DEM as an input. Results show that the hydraulic simulation based on the generated VHR-DEM, calibrated and validated via field data, produces an accurate extent and water level of the flooded area, proving that Remote Sensing stereo reconstruction is a promising alternative to the traditional survey methods in flood mapping applications.