Benthic Cover from Automated Annotation of Benthic Images Collected at Coral Reef Sites in the Pacific Remote Island Areas and American Samoa in 2018

The coral reef benthic community data described here result from the automated annotation (classification) of benthic images collected during photoquadrat surveys conducted by the NOAA Pacific Islands Fisheries Science Center (PIFSC), Ecosystem Sciences Division (ESD, formerly the Coral Reef Ecosystem Division) as part of NOAA's ongoing National Coral Reef Monitoring Program (NCRMP). SCUBA divers conducted benthic photoquadrat surveys in coral reef habitats according to protocols established by ESD and NCRMP during the ESD-led NCRMP mission to the islands and atolls of the Pacific Remote Island Areas (PRIA) and American Samoa from June 8 to August 11, 2018. Still photographs were collected with a high-resolution digital camera mounted on a pole to document the benthic community composition at predetermined points along transects at stratified random sites surveyed only once as part of Rapid Ecological Assessment (REA) surveys for corals and fish and permanent sites established by ESD and resurveyed every ~3 years for climate change monitoring. Overall, 30 photoquadrat images were collected at each survey site.

The benthic habitat images were quantitatively analyzed using the web-based, machine-learning, image annotation tool, CoralNet (https://coralnet.ucsd.edu; Beijbom et al. 2015). Ten points were randomly overlaid on each image and the machine-learning algorithm "robot" identified the organism or type of substrate beneath, with 300 annotations (points) generated per site. Benthic elements falling under each point were identified to functional group (Tier 1: hard coral, soft coral, sessile invertebrate, macroalgae, crustose coralline algae, and turf algae) for coral, algae, invertebrates, and other taxa following Lozada-Misa et al. (2017). These benthic data can ultimately be used to produce estimates of community composition, relative abundance (percentage of benthic cover), and frequency of occurrence.

The use of the CoralNet machine-learning, image annotation tool is essential to the timely and cost-effective analysis of benthic imagery collected Pacific-wide under the auspices of the NOAA National Coral Reef Monitoring Program (NCRMP). Timely analysis and reporting of cover estimates are fundamental to understanding status and change of coral reef benthic communities, particularly in a fast-changing climate.

The NOAA National Coral Reef Monitoring Program (NCRMP) details a long term approach to provide an ecosystem perspective via monitoring climate, fish, benthic, and socioeconomic variables in a consistent and integrated manner. The NCRMP coordinates various NOAA Coral Reef Conservation Program (CRCP) biological, physical, and human dimensions activities into a cohesive NOAA-wide effort. Through the implementation of the NCRMP, NOAA is able to clearly and concisely communicate results of national-scale monitoring to national, state, and territorial policy makers, resource managers, and the public on a periodic basis.

NCRMP is a framework for conducting sustained observations of biological, climate, and socioeconomic indicators at some ~40 priority coral reefs systems across the U.S. and its territories. This integrated approach consolidates monitoring of coral reefs under a uniform method in the Pacific, Atlantic, Caribbean, and the Gulf of Mexico for the first time. NCRMP is funded by the CRCP and supported by NOAA Fisheries, NOAA National Centers for Coastal Ocean Science (NCCOS), and many other partners. The Ecosystem Sciences Division at NOAA Fisheries is leading biological and climate monitoring in the U.S. Pacific Islands Region.

The biological component of NCRMP in the Pacific provides a triennial ecological characterization at a broad spatial scale of general reef condition for reef fishes, corals and benthic habitat (i.e., fish species composition/density/size, benthic cover, and coral density/size/condition). The climate component of NCRMP in the Pacific provides a comprehensive view of climate change impacts on coral reef ecosystems and helps identify areas of resilience and vulnerability. The key indicators used to identify and monitor climate-driven trends include 1) thermal stress caused by changes in sea temperature, 2) ocean acidification resulting from changes in carbonate chemistry, and 3) ecological impacts by collecting data on coral growth rates, erosion, and community structure to understand the impacts of thermal stress and ocean acidification on the ecosystem. Each year, ESD scientists work closely with CRCP and partners during Pacific RAMP missions to collect data using moored oceanographic (subsurface temperature recorders) and ecological (calcification accretion units [CAUs] and autonomous reef monitoring structures [ARMS]) instruments stationed at fixed sites in the Pacific Ocean, and water samples collected by divers. The in-situ data (along with satellite-based observations) are also used in modeling efforts. Innovative analysis techniques are used to develop products that give fellow scientists, managers, decision makers and the public a better understanding of a region's resources and how they are changing over time.

Created September 2019 for benthic data generated by robot from images collected in 2018.

Raw benthic community data include metadata for each survey (where, when, who); site characteristics (depth, reef zone); unique image name, and individual point observations of benthic elements identified to benthic taxa group, referred to as "Tier 1", which includes hard coral, soft coral, macroalgae, turf algae, crustose coralline algae, invertebrates, mobile fauna, and sediment.

While every effort has been made to ensure that these data are accurate and reliable within the limits of the current state of the art, NOAA cannot assume liability for any damages caused by errors or omissions in the data, nor as a result of the failure of the data to function on a particular system. NOAA makes no warranty, expressed or implied, nor does the fact of distribution constitute such a warranty.

Pacific Remote Island Areas (PRIA) including Baker, Howland, and Jarvis islands, Kingman Reef, and Palmyra atoll, and the islands and atolls of American Samoa.

Not applicable

Not applicable

NOAA Coral Reef Ecosystem Program (CREP) Data Sharing Recommendations, version 9.0 updated August 12, 2015:

CREP welcomes the opportunity to collaborate on research issues contributing to the scientific basis for better management of marine ecosystems. CREP has a very diverse set of field activities that generates large volumes of data using an array of data collection protocols.

The following recommendations are for your consideration as you use this data:

1) Data analyses should take all field exigencies into account. The most effective way to do this would be active collaboration with CREP principal investigators.

2) In all presentations, product releases, or publications using data generated by CREP, proper acknowledgement of both CREP and the individuals responsible for data collection is expected. Citing the DOI (if available) is preferred, a non-DOI example is listed below.

3) If you collect or generate data for the same study areas, CREP requests that you share relevant information on complimentary data collections.

4) Those receiving data are strongly urged to inform the CREP Data Management Team of any errors and discrepancies that are discovered during the course of using these data. They are further urged to bring to the attention of the Team all problems and difficulties encountered in using these data. This information is necessary in order to improve the collections and to facilitate more efficient and economical data processing and retrieval. The users are asked to supply copies of any missing data that may be located, and to provide information as to significant subsets and special aggregations of data that are developed in using the material provided.

Example citation:

"This publication makes use of data products provided by the Ecosystem Sciences Division (ESD), Pacific Islands Fisheries Science Center (PIFSC), National Marine Fisheries Service (NMFS), National Oceanic and Atmospheric Administration (NOAA), with funding support from the NOAA Coral Reef Conservation Program (CRCP). The analysis and interpretations presented here are solely that of the current authors.”

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

None

Please cite PIFSC Ecosystem Sciences Division (ESD) when using the data.

Example:

Ecosystem Sciences Division; Pacific Islands Fisheries Science Center, 2019: Benthic Cover from Automated Annotation of Benthic Images Collected at Coral Reef Sites in the Pacific Remote Island Areas and American Samoa in 2018, https://www.fisheries.noaa.gov/inport/item/57610.

None

None

PIFSC Oracle database view: CoralNet: MV_BIA_CNET_ANALYSIS_DATA_ROB

Data sets derived from the machine-learning, image annotation tool, CoralNet, are highly reliable within the stipulated taxonomic context (Tier 1). Substantial effort is required to manually annotate benthic images, which is expensive and leads to lags before data are available.

Human, expert-analyzed imagery from NOAA’s Pacific Reef Assessment and Monitoring Program is used to train the machine-learning image analysis tool, CoralNet (Beijbon et al 2015), to generate fully-automated Tier 1 benthic estimates of site-level coral cover that are highly comparable to those generated by human analysts (Pearson’s r > 0.97, and with bias of 1% or less; Williams et al. 2019). CoralNet was generally effective at estimating coral cover (Pearson’s r > 0.92 and with bias of 2% or less in 6 of 7 cases), but performance was mixed for other groups including algal categories.

As image acquisition is relatively straightforward, the capacity of fully-automated image analysis tools to minimize the need for resource-intensive human analysts creates the opportunity to dramatically increase the quantity and consistency of coral reef benthic data available to researchers and managers.

Handheld GPS units were used to locate and mark site locations.

Permanent transects/plots were marked to allow direct, site-level temporal comparisons.

The data set derived from the annotated images is as good as the images themselves; poor images with low contrast, inappropriate white balance, or out of focus may have been discarded. Therefore, total annotations for some sites may have differing number of points between sites (300 annotations are expected per site if there were 30 photographs for the site and 10 points analyzed per photograph). Although the analysis process is statistically robust to allow for differing n (points) between sites, users should be aware of this limitation.

Only shallow water (0 to 30 meter depth range) hard-bottom habitats are surveyed.

Classification is complete to the functional/taxa group level (Tier 1).

Survey sites in which less than 15 images were collected are discarded from the analysis.

The same method of image analysis was used at each surveyed site.

Machine-generated fully-automated benthic estimates of site-level coral cover (Tier 1) are highly comparable to those generated by human analysts (Pearson’s r > 0.97, and with bias of 1% or less; Williams et al. 2019).

Pre- and post-automated benthic image analysis quality control procedures include spot checks for:

1- Image quality integrity: focus, white balance, distance from bottom (too high or too low)

2- image inventory completeness (all images acquired for each island are included in the analysis)

3- Duplicate sites

4- Complete and coherent upload of imagery to CoralNet

5- Integrity and completeness of robot-generated image annotations

The image analysis data is captured in two different locations: exported files from CoralNet that reside on a file server, which are ingested into a table in the PIFSC Oracle database. Both the file server and PIFSC Oracle database are maintained and regularly backed up by PIFSC ITS.

Protocol describing the automated annotation of benthic images to generate benthic community data. Still photographs (30 per survey site) collected during photoquadrat surveys at random and permanent sites in coral reef habitats by the PIFSC Ecosystem Sciences Division were later annotated by a human-trained robot in CoralNet.