Imaging and 3D model of the substratum
We digitally reproduced ~ 0.25 m2areas of the reef, by extracting images from video recordings and applying SfM algorithms. We delimited each area with a black PVC quadrat (50 x 50 cm) leveled on the bottom. In some cases, macroalgae such asDyctiota spp. or Lobophora spp. covered up to 80% of the substrate and we manually removed the algae. We used a NIKON COOLPIX-AW100 with illumination from two 2500 lumen lights (Sola Video 2500) (to film high-definition videos (1920 x 1080p / 30fps) of each quadrat). We used a cross level attached to the quadrat (4.2 x 1.6 cm) to scale and orient the model parallel to the seawater surface (Supplementary Methods Figure 1). To generate accurate 3D reconstructions, we attached 12 ground control points (GCPs) to the quadrat (Supplementary Methods Figure 1). To capture each scene in hundreds of pictures with different focal distances and 70% overlap, a SCUBA diver filmed each area following a spiral pattern, starting 2 meters over the reef and finishing at ~ 50 cm (Young et al. 2017). Finally, we extracted three high-definition images per second from each video using the open-source software FFmpeg (www.ffmpeg.org), which yielded to 500 – 700 images per quadrat.
We built 3D dense point clouds and polygon meshes with Agisoft Metashape v1.6. The workflow to obtain dense point clouds, meshes and DEMs has been described in detail in previous studies (Burns et al. 2015; Bayley & Mogg 2020). Modifications and full details can be found in Supplementary Methods: Table 1. To obtain the coordinates of scleractinian and octocoral recruits, we marked and labeled the position of each recruit in the point cloud. Next, we manually annotated the point cloud based on substratum type, differentiating recruits (scleractinians ≤ 4 cm wide, octocorals ≤ 5 cm height) from calcareous rock, igneous rock sand and benthic invertebrates (scleractinian coral, octocoral base, and sponge). We excluded the PVC quadrat and the area beyond each quadrat from the analyses.
Our approach differs from most 2.5D methods, which use digital elevation models (DEMs) to study topography. Instead, we calculated slope and roughness directly from the point cloud at scales of 5, 10, 20, 50 and 100 mm. We also developed a new metric, the Topographic Exposure Index (TEI), which quantifies the degree of exposure to the surrounding environment. This metric distinguishes between flat, concave and convex topographic features in the 3D models at mm/cm grains. TEI is the orthogonal distance from each point of the cloud to each point of a smoother version of the same cloud (expressed in m). Positive values are assigned to points above the smoother surface (more exposed, such as those on a ’hill’), and negative values to those that fall below it (less exposed, such as those in a ‘hole’). The TEI is defined relative to the surrounding landscape, and a location with a positive TEI (a ‘hill’) could potentially be ‘pointed’ towards the water surface, or horizontally, or even directly down if it is located on the underside of an overhanging area. Slope, roughness, and the smooth cloud used to calculate TEI were calculated with CloudCompare v2.9 (see Supplementary Methods Table 2). The code for calculating TEI is written in R and available at github.com/AdamWilsonLab/meshSDM.