Qualitative and quantitative analysis of Pléiades and World-View 3 high resolution DEMs

Abstract

Recently, generating Digital Elevation Models (DEMs) from high-resolution stereo satellite imagery has gained significant interest as a research topic. The sub-meter Ground Sampling Distance (GSD) offered by these images suggests that decimeter-level accuracy can be achieved. The Pléiades and WorldView-3 sensors, advanced optical Earth observation satellites, are capable of capturing multiple perspectives of the same location in a single pass, providing valuable data for 3D modeling and extraction. Photogrammetrically derived DEMs are essential for applications in engineering, land planning, geomorphology, forestry, infrastructure, urban modeling, change detection, disaster management, and city planning. To ensure reliable analyses for the various applications, high quality elevation models are required. Therefore, the primary focus of this research investigation is to analyse and evaluate the geometric quality potential of DEMs derived from high-resolution Pléiades and WorldView-3 stereo and tri-stereo satellite imagery. The study area located in Allentsteig, Lower Austria is characterized by a hilly landscape with arable land and coniferous forests, and elevations ranging from 300 to 690 meters a.s.l. The entire photogrammetric workflow begins with VHR stereo/tri/stereo satellite imagery and includes the following processing steps: image orientation with improved Rational Polynomial Coefficients (RPCs), dense image matching, and 3D reconstruction using the forward intersection technique. The resulting 3D point clouds serve as input for deriving high resolution Digital Surface Models, whose accuracy is evaluated using reference Ground Control and Check Points (GCPs, CPs) and a LiDAR Digital Terrain Model (DTM). The vertical accuracy shows RMSE values of 0.96 m and of 0.37 m for Pléiades and WorldView-3 DSMs, respectively. To improve this outcome, the photogrammetrically derived DSMs were further aligned to the reference lidar DTM by applying an affine 3D transformation using the Least Squares Matching Technique (LSM). The results indicate improved vertical accuracy of 0.61 meters for Pléiades tri-stereo scenes and 0.24 meters for WorldView-3 tri-stereo scenes, respectively.