Structure-from-motion (SfM) algorithms greatly facilitate the production of detailed topographic models from photographs collected using unmanned aerial vehicles (UAVs). However, the survey quality achieved in published geomorphological studies is highly variable, and sufficient processing details are never provided to understand fully the causes of variability. To address this, we show how survey quality and consistency can be improved through a deeper consideration of the underlying photogrammetric methods. We demonstrate the sensitivity of digital elevation models (DEMs) to processing settings that have not been discussed in the geomorphological literature, yet are a critical part of survey georeferencing, and are responsible for balancing the contributions of tie and control points. We provide a Monte Carlo approach to enable geomorphologists to (1) carefully consider sources of survey error and hence increase the accuracy of SfM-based DEMs and (2) minimise the associated field effort by robust determination of suitable lower-density deployments of ground control. By identifying appropriate processing settings and highlighting photogrammetric issues such as over-parameterisation during camera self-calibration, processing artefacts are reduced and the spatial variability of error minimised.
We demonstrate such DEM improvements with a commonly-used SfM-based software (PhotoScan), which we augment with semi-automated and automated identification of ground control points (GCPs) in images, and apply to two contrasting case studies – an erosion gully survey (Taroudant, Morocco) and an active landslide survey (Super-Sauze, France). In the gully survey, refined processing settings eliminated step-like artefacts of up to ~50 mm in amplitude, and overall DEM variability with GCP selection improved from 37 to 16 mm. In the much more challenging landslide case study, our processing halved planimetric error to ~0.1 m, effectively doubling the frequency at which changes in landslide velocity could be detected. In both case studies, the Monte Carlo approach provided a robust demonstration that field effort could by substantially reduced by only deploying approximately half the number of GCPs, with minimal effect on the survey quality. To reduce processing artefacts and promote confidence in SfM-based geomorphological surveys, published results should include processing details which include the image residuals for both tie points and GCPs, and ensure that these are considered appropriately within the workflow.
This is the author’s version of a work that was accepted for publication in Geomorphology. Changes resulting from the publishing process, such as peer review, editing, corrections, structural formatting, and other quality control mechanisms may not be reflected in this document. Changes may have been made to this work since it was submitted for publication. A definitive version was subsequently published in Geomorphology, 280, 2017 DOI: 10.1016/j.geomorph.2016.11.021