Colony-level annotations using Structure-from-Motion models from sites surveyed across the Main Hawaiian Islands during the 2019 bleaching event

Main Hawaiian Islands (MHI), including Hawaii, Maui, Oahu, and Lanai.

MHI 2019 bleaching surveys

Description of fixed sites surveyed in 2019 using SfM methods to generate models that were annotated to derive coral colony patch level information.

Coral colony patch level SfM annotations observations from surveys conducted during the 2019 bleaching event in the Hawaiian Archipelago at fixed sites.

Reference table of full taxa names per taxa code.

This study utilized SfM 3D reconstruction software tools to create textured mesh models of a reef at French Frigate Shoals, an atoll in the Northwestern Hawaiian Islands. The reconstructed orthophoto and digital elevation model were then integrated with geospatial software in order to quantify metrics pertaining to 3D complexity. The resulting data provided high-resolution physical properties of coral colonies that were then combined with live cover to accurately characterize the reef as a living structure. The 3D reconstruction of reef structure and complexity can be integrated with other physiological and ecological parameters in future research to develop reliable ecosystem models and improve capacity to monitor changes in the health and function of coral reef ecosystems.

This study provides the first quantified methodological comparison to validate the transition from standard in-water methods to SfM survey methods for estimates of coral colony-level surveys. Here, we quantitatively compare data generated from in-water surveys to SfM-derived metrics for assessing coral demography, bleaching, and diversity in the main Hawaiian Islands as part of NOAA’s National Coral Reef Monitoring Program. Our objectives were to compare between-method error to within-method

error, test for bias between methods, and identify strengths and weaknesses of both methods. Colony density, average colony diameter, average partial mortality,

prevalence of bleaching, species richness, and species diversity were recorded using both methods within the same survey areas. For all metrics, the magnitude of between-method error was comparable to the within-method error for the in-water method and between method error was significantly higher than within-method error for SfM for

one of the seven metrics. Our results also reveal that a majority of the metrics do not vary significantly between methods, nor did we observe a significant interaction between method and habitat type or method and depth. Exceptions include estimates of partial mortality, bleaching prevalence, and Porites juvenile density–though differences

between methods are generally small. Our study also highlights that SfM offers a unique opportunity to more rigorously quantify and mitigate inter-observer error by providing observers unlimited “bottom time” and the opportunity to work together to resolve difficult annotations.

The combination of 3D acquisition (terrestrial and airborne LiDAR, structured light, structure-from-motion) and 2D imaging (photographic, multispectral, panoramic, orthorectified, reflectance transformation) techniques allows the geometry, appearance and other aspects sites to be objectively documented. This study proposes a system for the direct visualization and analysis of such data, allowing the different aspects recorded to be layered together, and co-visualized with annotations and other relevant information. This study describes the required technical foundations, including gigapoint and gigapixel visualization pipelines that enable the dynamic layering of high-resolution imagery over massive minimally-processed LiDAR point clouds that serve as the base spatial layer. In particular, the study introduces the pointbuffer—a GPU-resident view-dependent point cache—as the foundation of our gigapoint pipeline, and outline the use of virtual texturing for draping of gigapixel imagery onto point clouds.

NOAA technical memorandum NMFS PIFSC

Here, we outline a method of directly measuring coral vital rates via structure-from-motion (SfM) photogrammetry — an image processing technique capable of reconstructing accurate, orthorectified representations of the seafloor and the coral communities that inhabit them. By leveraging SfM technology, NOAA’s U.S. National Coral Reef Monitoring Program (NCRMP) and partners are capable of generating accurate measures of coral vital rates from the scale of individual coral colonies with coverage of tens of sites and thousands of corals within a population.

In Section 1 of this technical memorandum, we provide a detailed set of Standard Operating Procedures for capturing, processing, and extracting data from SfM photogrammetry. In Section 2, we use this novel SfM approach and imagery collected at 14 sites across the Hawaiian Archipelago across time intervals from a few minutes to three years to determine how accurately we can measure coral vital rates.

NOAA technical memorandum NMFS PIFSC

This document provides detailed procedures for collecting and processing imagery using Structure-from-Motion techniques developed by Ecosystem Sciences Division (ESD) in collaboration with Scripps Institution of Oceanography and the University of Hawaii at Hilo. These procedures are designed to efficiently generate coral demographic and benthic community metrics across the broad spatial scale of the Pacific Rapid Assessment and Monitoring Program. This pipeline consists of four key steps: (1) Image collection by SCUBA divers, (2) Data management, post-processing, and QC, (3) Generating 3-D models and 2-D orthophotos in Agisoft Metashape, and (4) Extracting demographic data in ArcMap. This SOP is the result of comprehensive testing of different camera systems, collection techniques, and software. While the following procedures are designed to meet ESD needs, we primarily use commercially available cameras and software, making these methods adaptable based on programmatic capacity and needs.

Guidelines and procedures for implementation of sampling design, in situ survey methodology, and data entry for the monitoring of reef coral populations and benthic communities as part of the Pacific RAMP led by ESD (Swanson et al. 2018).

NOAA technical memorandum NMFS PIFSC

The objective of this document is to provide updates incorporated in 2019 to the guidelines and procedures for implementation of sampling design, survey methodology, and data entry for the monitoring of reef coral populations and benthic communities as part of the Pacific RAMP led by ESD (Swanson et al. 2018).

SfM benthic surveys were collected at fixed sites during the 2019 bleaching event in the Main Hawaiiian Islands. A 10x10m or 12m diameter circular plot was visually established. 3-4 ground control points (GCPs) were placed within the plot for scale. White balance settings were adjusted in situ using a 15% gray card. JPEG images were collected using a Nikon SL2 digital camera in an underwater housing with a dome port. Images were collected at each site by swimming in a back-and-forth or circular motion 1m above the substrate capturing images continuously to achieve a 60-80% overlap.

Images for each site were evaluated for image quality and images deemed unsatisfactory (e.g. overexposed, images of blue water or images of divers, or images not taken perpendicular to the reef) were removed from the image set.

The Structure from Motion (SfM) approach produces an accurately scaled, two-dimensional (2D) orthomosaic model created from the overlapping imagery. Raw imagery was imported into Agisoft Metashape software where images were aligned and used to build 3D dense point clouds (DPCs) following parameters described by Rodriguez et al. (2021). DPCs were then imported into Viscore, a visualization software (Petrovic et al., 2014) where they were scaled and oriented using the GCP information. The ground sample distance (GSD) of the scaled DPC, which estimates the resolution per pixel by measuring the size of each pixel on the ground, ranged from 2-4 mm/pix. A geometrically accurate 2D projection of the DPC (orthoprojection) and scale grid are exported from Viscore and uploaded into ArcGIS Pro for annotation.

In ArcGIS Pro, each site was set up for annotation by scaling the orthoprojection using the scale grid exported from Viscore, manually digitizing quadrats as a shapefile, and setting up the attribute table in a geodatabase. To record and extract data from the orthoprojection image, each coral colony patch >= 5cm was annotated. Each patch was measured by digitizing a line across the maximum diameter of the colony. Coral ID (to lowest taxonomic level), morphology, and bleaching extent and severity were recorded. During annotation, the original JPEG imagery was viewed alongside the orthoprojection with the Agisoft or Viscore Image View feature to see fine scale colony details, observe colonies from multiple angles and locate colonies not visible in the orthoprojection (e.g. under ledges).

Following Rodriguez et al. (2021), if a coral species per site was found less than 10 times in the first 10 quadrats, the species was subsequently not recorded in future analyzed quadrats. Therefore, this data should be carefully analyzed and species density should not be calculated for these species with low sample size. If a species had low density at a given site, 20-40 patches were annotated. If a species was abundant at the site, at least 40 patches were annotated. See Rodriguez et al. (2021) for further information.

Annotations created in ArcGIS Pro were quality controlled using a multi-stage process. Data were exported from ArcMap and quality controlled in R with specific queries to identify and correct data entry errors (e.g. misspelled species names, data in incorrect columns, bleaching >100%).