We use a four-step large-area imaging protocol to collect ecological data for structural complexity, community composition, and demographic analyses of benthic marine ecosystems. In each new application, the biggest challenges are defining the needed resolution in the raw imagery, determining the spatial extent of the area to be imaged, and ensuring that we have adequate plot-level replication for accurate scientific analysis. This protocol emphasizes the value of the source imagery throughout the four-step process, ensuring that quality images are collected, archived, and used to facilitate detailed ecological data extraction for analyses.
These data handling and visualization workflows, especially those that utilize the raw imagery, ensure compatibility between data collected digitally in the laboratory or by scuba divers in the field. This, in turn, allows integration of these digitally enhanced approaches into existing long-term datasets. This approach allows for a dramatic increase in the spatial extent in replication of the data we collect, allowing us to ask spatially explicit questions and conduct more robust demographic analyses.
Most importantly, it increases our capacity for tracking ecological change through time. To begin, attach outer frame panels of the camera frame to camera mounting panels and columns using 1 1/2-inch long Phillips flathead screws. Prepare two DSLR cameras with one equipped with a fixed wide angle lens and the second equipped with a zoom lens.
Attach and secure the dome port to assemble the underwater camera housings. Then attach the handles with 1/2 an inch long Phillips head screws. Fix the camera mounting plate using a 1 1/8-inch long socket head screw.
Next, insert the cameras into the housing. And use the vacuum pump to set the housing pressure to five inches of mercury to verify the integrity of the O-ring seal. Now, slide the camera mounting plate onto the mounting frame panels to install the housings onto the camera frame.
Secure the housings in place with thumb screws. For image capture, start each camera on an intervalometer set to capture at a rate of one frame per second. Swim the camera system approximately 1.5 meters above the benthos in a gridded pattern.
Perform a second gridded pass perpendicular to the first, maintaining approximately one meter spacing between each pass. Ensure the passes extend a minimum of two meters beyond the plot boundaries to ensure sufficient overlap within the target plot area. Launch the software for image processing on a computer system.
Click on Workflow, followed by Add Folder to load all the images into the Agisoft Metashape project. Once the files have loaded, select the data layout as Single cameras, Add all images to one chunk. Remove images with excessive blue water in the scene.
Now, click on Workflow, followed by Align Photos to align all the images. Verify that the image set has successfully aligned by checking the percentage of cameras aligned. Inspect the generated sparse point cloud for gaps in coverage or misalignments.
Ensure the bounding box encompasses the entire sparse point cloud before proceeding. Modify the bounding box if necessary using the Resize or Rotate Region options. Next, disable the camera group containing the zoom lens images.
Construct the dense point cloud by selecting Workflow, following by Build Dense Cloud. Sequentially click on Tools, Run Script, Extract Meta PY Script to export the camera pose estimates. Then click on File, followed by Export and Export Points to export the dense point cloud.
Drag and drop the exported dense point cloud file onto the vc5prep-confident. bat file located in the visualization software's program files. Compile the exported data files, including the camera pose files, along with the generated program files, into a single directory for use in the visualization software.
Use the rugo tool to create a 10-meter by 10-meter box on the dense point cloud. Set the maximum dimension to 10 meters and the aspect ratio to 1.0 to designate the 100-square-meter target area for data extraction. Next, use the cams tool to link the source images to the dense point cloud.
Enable spatially queried multi-image views of points on the model. For a density survey, after the images have been linked to the software, change the focal length of the perspective view to 100 millimeters to set a pseudo map view of the dense point cloud. Zoom out to a top-down, full view of the model.
Now, use the given quadrant sampling file to capture the view in the web applet by clicking eval for cell C1 and selecting the grab button. Turn on cams and link images within the quadrant sampling workflow by clicking eval for cells C2 and C3 in the quadrant sampling script. Turn on the previously-made rugo box for the 100-square-meter data extraction area.
In the web applet, eval the C4 prep cells section to sample 100 quadrants of one square meter each. In the quadrant sampling web address, use source imagery to search through a quadrant. Use a double left-click to retarget the sampling location and click a taxonomic button to designate the targeted point as a sample.
To remove a marked point, double left-click and select nothing. Compile all sampling files located under asterisk aux recruits test1 into a single directory. Then rename each file to include the site name.
Add the button lookup file to the directory. Run the onscreen script following inline instructions to aggregate the sample data by site and taxonomic group. To prepare data for submission to a repository, generate a methods description file that includes survey details, such as the area covered, camera system, ground control markers, and collection pattern.
Then generate a survey metadata file specific to the image dataset, including fields such as site name, collection data, GPS coordinates, plot bearings, ground control depth, and scale data, and the collection pattern and camera system used. Combine the description file, metadata file, and image files into a single ZIP archive for ingestion into the data repository. Successful large-area image collection resulted in the creation of a dense point cloud reconstruction with full top-down coverage of the survey area, while inadequate redundancy in coverage resulted in gaps or full degradation of the point cloud.
Measures of linear rugosity extracted from large-area image, or LAI, surveys closely aligned with in situ measures of complexity across sites, apart from outliers. Benthic community composition and percent coverage of functional groups from LAI matched those from traditional photo quadrat surveys. Sessile invertebrates, particularly sea urchin abundance, recorded using LAI methods was consistently higher than in situ methods due to comprehensive area coverage in LAI surveys.
Coral colony segmentation using LAI surveys revealed similar size distributions of common coral taxa compared to in situ methods. Co-registration of dense point clouds allowed monitoring of reef changes over time, even in dynamic environments with high growth and structural loss, as demonstrated at Millennium Atoll.
