Improving Satellite Stereo Photogrammetry Through Calibration With Space Borne LiDAR: A Case Study in The Netherlands

By Steven van Ree

Abstract
During the last decades very high resolution (VHR) satellite imagery has become commonplace. This has supported the development of open-access stereo photogrammetry tools and pipelines for the generation of Digital Surface Models (DSMs). These software are now capable of Structure from Motion and full Multi-View Stereo methods, adopted from the computer vision domain. However, the Rational Polynomial Coefficient (RPC) model used to relate satellite image data to real world coordinates results in limited positional accuracy in DSM construction. In this research DSM calibration with space borne LiDAR is proposed to improve the DSM accuracy further. Three satellite stereo photogrammetry pipelines are implemented: ASP, CARS and SSR-COLMAP. The DSMs are subsequently calibrated with space borne LiDAR from the ICESAT-2 ATLAS instrument, which has global coverage and sub-meter elevation accuracy. It was found that SSR-COLMAP performed best overall, with a mean error of 5.05m, which improved by 90% to 0.50m after calibration. The RMSE improved from 8.49 to 5.55m (35% improvement). Additionally, analysis of satellite imagery metadata proved that stereo photogrammetry can be optimized by excluding many poor performing image pairs. In general, a short acquisition period (<2months, independent of year) and small difference in solar zenith angle (<15 degrees) are related to higher-quality DSMs. A small satellite baseline (B/H 0.1 – 0.3) is also an indicator for successful DSM generation in an urban setting, while in rural areas the baseline should be greater (B/H 0.2 – 0.5) for better results.

Keywords: DSM; LiDAR; calibration; stereo photogrammetry; VHR