Reducing systematic dome errors in digital elevation models through better UAV flight design

It is well established that digital elevation models (DEMs) derived from unmanned aerial vehicle (UAV) images and processed by structure from motion may contain important systematic vertical errors arising from limitations in camera geometry modelling. Even when significant, such 'dome'-shaped errors can often remain unnoticed unless specific checks are conducted. Previous methods used to reduce these errors have involved: the addition of convergent images to supplement traditional vertical datasets, the usage of a higher number of ground control points, precise direct georeferencing techniques (RTK/PPK) or more refined camera pre-calibration. This study confirms that specific UAV flight designs can significantly reduce dome errors, particularly those that have a higher number of tie points connecting distant images, and hence contribute to a strengthened photogrammetric network. A total of 22 flight designs were tested, including vertical, convergent, point of interest (POI), multiscale and mixed imagery. Flights were carried out over a 300 x 70 m(2) flat test field area, where 143 ground points were accurately established. Three different UAVs and two commercial software packages were trialled, totalling 396 different tests. POI flight designs generated the smallest systematic errors. In contrast, vertical flight designs suffered from larger dome errors; unfortunately, a configuration that is ubiquitous and most often used. By using the POI flight design, the accuracy of DEMs will improve without the need to use more ground control or expensive RTK/PPK systems. Over flat terrain, the improvement is especially important in self-calibration projects without (or with just a few) ground control points. Some improvement will also be observed on those projects using camera pre-calibration or with stronger ground control. (c) 2020 John Wiley & Sons, Ltd.