Large wood (LW) 3D accumulation mapping and assessment using structure from Motion photogrammetry in the laboratory

Abstract

A Structure from Motion (SfM) photogrammetry-based methodology for precise mapping of large wood (LW) accumulations in fluvial systems is presented. The technique may be useful for routine inventory and rapid volume estimation of complex LW structures that tend to divert or obstruct flow in rivers. Our methodology is validated by means of laboratory experiments, ranging from elementary to complex arrangements of LW components and organic fine material (OFM). Seven experimental setups were used as the basis for (i) manual approximations of 2.5D and 3D geometric reference volumes, (ii) 2.5D volume models based on mesh geometry, (iii) a closed (‘watertight’) 3D mesh from an unorganized point cloud and (iv) estimates of porosity. A commercially available SfM photogrammetry software package, Pix4Dmapper, was used for point cloud and simplified mesh generation. In order to obtain more precise volumes, 3D surface models are required. Accordingly, we generated 3D watertight mesh models of the unorganized point cloud using the screened Poisson Surface Reconstruction (PSR) technique. The Pix4D volume tool (2.5D) resulted in an overestimation (2.9 to 52.7%) of the geometric volume (3D), due to the convex hull approximation of the geometry normal to the surface plane. PSR (3D) resulted in more precise volumes, showing deviations from the geometric volume (3D) in a range of −15.9 to +10.6%, as the algorithm could capture concavities and involuted surfaces on the accumulation. Assuming that the difference between 3D and 2.5D volume models represents many of the voids visible within or on the surface of the deposit, this volume can be used as an estimate of porosity. Our assessment suggests that image-based SfM methodology is well-suited for further investigations in LW research, due to its time and cost efficiency in comparison with other conventional surveying techniques. The methodology can be used to generate high quality point cloud and mesh models of log jam formations; these may be used by river managers and researchers to quantify accumulation volume and thus gain a better understanding of LW composition and the influence of wood geometry on hydraulic flow conditions.