From experimental plots to experimental landscapes: topography, erosion and deposition in sub-humid badlands from Structure-from-Motion photogrammetry

In the last decade advances in surveying technology have opened up the possibility of representing topography and monitoring surface changes over experimental plots (<10 m2) in high resolution (~103 points m-1). Yet the representativeness of these small plots is limited. With ‘Structure-from-Motion’ (SfM) and ‘Multi-View Stereo’ (MVS) techniques now becoming part of the geomorphologist's toolkit, there is potential to expand further the scale at which we characterise topography and monitor geomorphic change morphometrically. Moving beyond previous plot-scale work using Terrestrial Laser Scanning (TLS) surveys, this paper validates robustly a number of SfM-MVS surveys against total station and extensive TLS data at three nested scales: plots (<30 m2) within a small catchment (4710 m2) within an eroding marl badland landscape (~1 km2). SfM surveys from a number of platforms are evaluated based on: (i) topography; (ii) sub-grid roughness; and (iii) change-detection capabilities at an annual scale. Oblique ground-based images can provide a high-quality surface equivalent to TLS at the plot scale, but become unreliable over larger areas of complex terrain. Degradation of surface quality with range is observed clearly for SfM models derived from aerial imagery. Recently modelled ‘doming’ effects from the use of vertical imagery are proven empirically as a piloted gyrocopter survey at 50m altitude with convergent off-nadir imagery provided higher quality data than an Unmanned Aerial Vehicle (UAV) flying at the same height and collecting vertical imagery. For soil erosion monitoring, SfM can provide data comparable with TLS only from small survey ranges (~5 m) and is best limited to survey ranges ~10–20 m. Synthesis of these results with existing validation studies shows a clear degradation of root-mean squared error (RMSE) with survey range, with a median ratio between RMSE and survey range of 1:639, and highlights the effect of the validation method (e.g. point-cloud or raster-based) on the estimated quality.