Data Management Plan (Deprecated)

The data described here are structural complexity estimates derived from Structure-from-Motion (SfM) benthic habitat imagery from diver transect surveys conducted at stratified-random sites (StRS) by the Pacific Islands Fisheries Science Center (PIFSC) Ecosystem Sciences Division (ESD) as part of the NOAA National Coral Reef Monitoring Program (NCRMP). The coral reef photos, collected by scuba divers, are used to generate 3-dimensional models, digital elevation models (DEMs) and orthomosaics using Agisoft Metashape photogrammetric software. The digital elevation models were analyzed in R studio to derive estimates of habitat structural complexity, using metrics such as: rugosity, fractal dimension, and height range, among others. The source imagery is archived and described separately at https://www.fisheries.noaa.gov/inport/item/63095.

The SfM surveys are part of the Rapid Ecological Assessment (REA) surveys for corals and fish conducted during ESD-led NCRMP missions. Sites were imaged using underwater cameras along 3x20m (3x13m for sites deeper than 18m) transects across the U.S. Pacific Island areas beginning in 2019. Please refer to the Lineage section for SOPs -- 2019 and 2022 have slightly different processing steps. Complexity metrics were estimated for each site within five or six 4m2 DEM sampling plots, which can then be averaged for the site. Rugosity is measured as the ratio between surface area and planar area within a sampling plot, Fractal dimension is measured using the variation or standard deviation method (see Lineage section below), and height range is measured as the difference between the deepest and shallowest point within a sampling plot.

A stratified random sampling (StRS) design was employed to survey the coral reef ecosystems across the U.S. Pacific Regions (Hawaiian Archipelago, Mariana Archipelago, American Samoa, and the Pacific Remote Island Areas). The survey domain encompassed the majority of the mapped area of reef and hard bottom habitats in the 0-30 m depth range. The stratification scheme included island, sub-island sector, reef zone, and depth in all regions, as well as habitat structure type. Allocation of sampling effort was proportional to the strata hardbottom area. The StRS design effectively reduces estimated variance by sampling more sites with larger hardbottom area and variance.

Extent of imagery used to generate the 2019 structural complexity dataset, at sites across the Main Hawaiian Islands (MHI), including Hawaii, Kauai, Maui, Oahu, Molokai, Niihau, Kahoolawe, and Lanai during the 2019 MHI NCRMP mission (SE1902_MHI_Leg3).

Extent of imagery used to generate SfM-derived complexity dataset, at sites across Guam and the Commonwealth of the Northern Mariana Islands (CNMI), together forming the Mariana Archipelago, during the 2022 Rainier Integrates Charting, Hydrography, and Reef Demographics (RICHARD) NCRMP mission (RA2201_Leg3-7).

Bounding coordinates that include US Pacific Islands region. More specific geographic information is included for 2019 and 2022 in this metadata record. More recent data are captured in these coordinates.

Lineage Statement:
The benthic SfM survey methodology, employed by the NOAA Ecosystem Sciences Division (ESD), began in 2019. Benthic imagery is collected at stratified random sites and generated into image maps using Agisoft Metashape software. Resulting digital elevation models are used to estimate three habitat structural complexity metrics: rugosity, fractal dimension, and height range.

Process Steps:

• A stratified random sampling (StRS) design was employed to survey the coral reef ecosystems through the U.S. Pacific regions. The survey domain encompassed the majority of the mapped area of reef and hard bottom habitats. The stratification scheme included island, reef zone, and depth in all regions, as well as habitat structure type in the Main Hawaiian Islands. The habitat structure types included simple, complex, and coral-rich. Depth categories of shallow (0-6 m), mid (> 6-18m) and deep (>18-30 m) were also incorporated into the stratification scheme. Allocation of sampling effort was proportional to strata area and variance in coral density. Sites were randomly selected within each stratum. A geographic information system (GIS) and digital spatial databases of benthic habitats (NOAA National Centers for Coastal Ocean Science NCCOS), reef zones (IKONOS satellite imagery, NDGC 1998) bathymetry (NDGC 1998, CREP benthic mapping data), and marine reserve boundaries (NOAA) were used to facilitate spatial delineation of the sampling survey domain, strata, and sample units. Map resolution was such that the survey domain could be overlain by a grid using a GIS with individual cells of size 50 m by 50 m in area. A one-stage sampling scheme following Cochran (1977) was employed. Grid cells containing at least 10% hard-bottom reef habitats were designated as primary sample units (referred to as sites).
• SfM benthic surveys were conducted by both the fish and benthic teams to record the benthic community composition. A 3 x 20 m plot was visually established centered on a transect line (at sites >18 m, a 3 x 13 m 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 or Canon SL3 digital camera in an underwater housing with a dome port. Images were collected at each site by swimming in a back-and-forth motion 1 m above the substrate capturing images continuously to achieve a 60-80% overlap. (Citation: Torres-Pulliza D, Charendoff J, Couch C, Suka R, Gray A, Lichowski F, Amir C, Lamirand M, Asbury M, Winston M, Basden I, Oliver T (2024). Processing Coral Reef Imagery Using Structure-from-Motion Photogrammetry: Standard Operating Procedures (2023 Update). https://doi.org/10.25923/CYDJ-Z260)
• 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. Images with compromised color tones (i.e. no white balance, dark, over exposed, etc.) were corrected using Adobe Lightroom. Image adjustments were batch run per site for consistent correction. Once the images have been collected, quality checked, and organized, they are processed to digitally reconstruct the 3D geometry of a coral reef scene through SfM using Agisoft Metashape. First, the raw imagery was imported into Agisoft Metashape Professional software where images were aligned and used to build 3D dense point clouds (DPCs) following parameters described by Torres-Pulliza et al. (2023). Two digital elevation models (DEM) were generated for each survey site and exported at a 0.002m resolution, one with default interpolation methods and one with interpolation methods disabled. Default interpolation was desired to acquire a DEM with no holes and was the product used for structural complexity metrics extraction. To avoid areas where limited data was interpolated, a non-interpolated DEM was extracted for use of the accurate reconstruction boundary. No analysis was performed on the non-interpolated DEM, but was used as a guide to trim the interpolated DEM. The interpolated digital elevation models were used to quantify coral reef surface complexity through metrics such as Rugosity, Fractal dimension and height range, among others. (Citation: Torres-Pulliza D, Oliver T, Weible R, Lichowski F, Kindinger T, Gajdzik L, Couch C (2025). Linking fish biomass and coral reefs structural complexity across the Mariana Archipelago (PIFSC Technical Memoranda Series, TM-PIFSC-180). Pacific Islands Fisheries Science Center.)
• In statistical software R v3.6.1, both digital elevation models (DEMs) were imported. Interpolated DEMs were clipped by the extent of the non-interpolated model to avoid inaccurate edges. Specifically, the non-interpolated DEM was aggregated with an aggregation factor of 100 to smooth the edges, and converted to a polygon. The resulting polygon was used to crop the interpolated DEM, thereby removing the distorted edges. Each belt was subsampled to more easily sample across belts that varied slightly in model straightness and width. Five or six 2 x 2m boxes were placed on each belt transect by randomly pulling coordinates from a normal distribution and only boxes that were fully within the bounding area were used (i.e., containing no NA values). Random boxes were continuously placed until five were achieved. Plots that were too thin for the 2 x 2m boxes were removed (14 sites). DEMs were spatially rectified to a known local reference system; however, the natural slope of some sites yielded inaccurate height range values (i.e., difference between maximum z and minimum z) when analyzing the subsetted sample box. To resolve this issue, tilt of sample boxes were corrected by updating the z values to reflect the residuals. Scripts modified from Torres-Pulliza et al. (2020) were used to calculate three structural complexity metrics: rugosity, height range, fractal dimension (variation method). The three metrics were extracted from each 2 x 2m box; Rugosity and height range were calculated at a resolution of 1 cm, while fractal dimension was calculated over a range of 1 cm to 2 m (thus over 2 orders of magnitude). Rugosity was calculated as the ratio between surface area and planar area. Height range was calculated as the difference between the deepest and shallowest point in the box. Fractal dimension was calculated using the variation method; each box was divided into squares with side lengths 2.00, 1.00, 0.64, 0.32, 0.16, 0.08, 0.04, 0.02, 0.01 m and height ranges within each grid at each scale were calculated and averaged for that scale. The slope was calculated for each box by fitting a line to the log of side lengths versus the log of mean height and fractal dimension was calculated by 3 - slope. See Torres-Pulliza et al. (2025) for an alternative sampling strategy and data extraction in R. (Citation: Asbury M, Schiettekatte, NMD, Couch CS, Oliver T, Burns JHR, Madin JS (2023). Geological age and environments shape reef habitat structure. Global Ecology and Biogeography 32, 1230-1240. https://doi.org/10.1111/geb.13691)

Prior to generating the 3D dense point clouds and 2.5D digital elevation model , the JPEG imagery was 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. Digital elevation models used in this analysis were extracted at 2mm resolution, which was within range of model accuracy listed above. During estimation of structural complexity in R Studio, scripts were controlled to place five or six 2 x 2m sampling plots on each belt transect by randomly pulling coordinates from a normal distribution and only boxes that were fully within the bounding area were used. Random sampling plots were continuously placed until the set number was achieved. Plots that were too thin for the 2 x 2m boxes were automatically removed. Data was secondly quality controlled in R and visually to ensure no data errors (e.g. missing values, erroneous values).

Data can be accessed online via the NOAA National Centers for Environmental Information (NCEI) Ocean Archive.

NOAA IRC and NOAA Fisheries ITS resources and assets.