Monitoring ephemeral river changes during floods with SfM photogrammetry

Abstract

Monitoring detailed topographic changes in rivers is an essential tool for understanding the morpho-sedimentary behavior of rivers. However, as the resolution of topographic reconstruction techniques improves, survey costs and the time consumed increase. In this paper, the emerging Structure from Motion Multi-View Stereo (SfM-MVS), a high resolution but low cost technique, is scrutinized to assess whether it constitutes a viable option for ‘change detection’ in ephemeral rivers. To that end, three photogrammetric flights were carried out to reconstruct the subsequent digital elevation models (DEMs) and quantify fluvial change caused by two floods in 2015 along a 6.5 km reach in an ephemeral gravel-bed river (Rambla de la Viuda; eastern Spain). The application of SfM-MVS resulted in a maximum point cloud density of 1295.4 pts/m2 and a DEM resolution of 3.11 cm/pix. Individual flights registered errors of 0.101, 0.171 and 0.083 m. Level of detection (LoD) of DEMs of difference (DoDs), i.e. topographic change detection, resulted in 0.198 and 0.190 between the first and second, and second and third flights respectively. Orthomosaics were also successfully created at a maximum resolution of 2.50 cm/pix. Analysis of the best configuration of SfM-MVS for ephemeral river monitoring indicated that a high overlap of photographs and, therefore, a large number of projections were critical for an efficient workflow and a high-quality model. To ensure model quality and survey efficiency, the following configuration is recommended: (1) overlap index of more than 20 projections, (2) flight distribution at two heights, and (3) the use of ground control points. SfM-MVS topographies and DoDs showed a discontinuous pattern represented by a succession of erosive-depositional sequences. Evolution of one of these sequences has been studied in detail and the legacy of a past mining pit was pointed out to be the principal driver for this morphosedimentary pattern. Change detection quantified a net erosion of (−)3118 m3 for Flood #1 and a net deposition of (+)787 m3 for Flood #2, at a 95% confidence. Knickpoint retreat, riverbed lowering and bank erosion were identified as the principal sources of sediment. Analyzing separately each flood highlighted that the mobilization of sediments was not proportional to peak discharges (98 and 80 m3/s, respectively). Interpretation of this behavior was hypothesized to be produced by the difference in total water volume (32.5 and 7.1 hm3), longer discharge period (24 and 11 days), different entrainment thresholds of each flood, or a clockwise hysteresis effect in the sediment transport, probably due to varying sediment availability. Thus, it was concluded that SfM-MVS worked especially well for change detection in ephemeral rivers and served as a basis for understanding morphological river patterns associated to floods and human impacts.