Automatic extraction and hydro-flattening of water bodies in digital surface and terrain models

Point clouds obtained from LiDAR (Light Detection and Ranging) and photogrammetric sensors allow the creation of a high-resolution Digital Elevation Models (DEM). DEM's directly generated from the point clouds, contains geometric irregularities that occur due to triangulation artifacts or inaccurate elevation information on water surfaces. The smooth representation of the water bodies on the DEM can be achieved with the hydro-flattening process performed using break lines drawn by manual methods that require a high amount of time and labor. In this study, an approach has been developed in the MATLAB environment to automatically detect/extract and hydro-flatten standing and stream water bodies on the DEM. The proposed approach uses the geometric information contained in the point clouds which are obtained from the digital aerial photographs (Rize, Malatya and Edirne) with dense matching and LiDAR sensors (Istanbul). The feature maps were generated with the 3D structure tensor over the point cloud and segmented using the SLIC (Simple Linear Iterative Clustering) algorithm. Then, the generated feature maps were classified into two classes, water and non-water, by the Random Forest classifier, and hydro-flattening operation was carried out over the detected water bodies on the DEM. F-Score, Correctness, Completeness and Quality metrics were used in the accuracy analysis step. The average values of the accuracy metrics calculated for each of the study areas which were divided into sub-data sets; 85,7%, 99,7%, 96,7% and 96,4% for Study Area-1 Sea (Rize), 93,7%, 99,6%, 95,9% and 95,5% for Study Area-2 (Malatya), 94,9%, 97,9%, 95,4% and 93,5% for Study Area-3 (Edirne) and for Study Area-4 (İstanbul), 73,6%, 95,3%, 94,3% and 90,1% were determined respectively. A 3D water surface was created using the water surface elevations obtained by intersecting water body boundaries with the Digital Terrain Model (DTM). Finally, the hydro-flattening operation was carried out by using this surface.