Guam Long-term Coral Reef Monitoring Program Benthic Cover Derived from Analysis of Benthic Images since 2010

The Government of Guam's Comprehensive Long-term Monitoring at Permanent Sites in Guam project, also known as the Guam Long-term Coral Reef Monitoring Program (GLTMP), is a NOAA-funded project currently coordinated through the University of Guam Marine Laboratory. The program involves the regular, intensive collection of data for a suite of coral reef ecosystem health parameters at high priority reef areas (HPRAs) around Guam, as well as critical support for coral bleaching response and other activities carried out by the multi-partner Guam Coral Reef Response Team. The program currently utilizes a split-panel sampling approach, whereby a mix of permanent and non-permanent sampling stations (one sampling station = one transect) are visited within each HPRA. The HPRAs were selected by an advisory body comprised of reef managers, researchers, and technicians; these reef areas were not intended to be representative of Guam's reefs more broadly. The locations of the sampling stations within each HPRA are generated randomly using GIS software. Various coral reef surveys are carried out on an annual basis along the seaward slope between 7 and 15 m depth in the Tumon Bay Marine Preserve and in East Agana Bay, while surveys are carried out biennially within the Piti Bomb Holes Marine Preserve, the Achang Reef Flat Marine Preserve, the eastern side of the Cocos Barrier Reef (Cocos-East), and Fouha Bay. Surveys were also carried at along reef margin (1-2 m) and slope (2-15 m) of Western Shoals, in Apra Harbor, in 2011. The surveys, which are currently conducted by University of Guam Marine Laboratory biologists, and which were supported by NOAA PIRO through 2019, currently include benthic photo transects, stationary point count fish surveys, macroinvertebrate belt transects, and chain-length rugosity surveys. The GLTMP has conducted surveys at the Tumon and East Agana HPRAs since 2010 and the Piti HPRA since 2012. Data collection for the Achang and Cocos-East HPRAs began in 2014 and at the Fouha Bay HPRA in 2015. Baseline data is available for the Western Shoals HPRA from 2011 but this site has not been re-visited since its establishment due to shifting management priorities.

The data products described herein document the status and trends for benthic cover and composition at the monitoring sites mentioned above. Benthic cover and coral and algal generic diversity are currently assessed using digital photo transects. Benthic cover estimates were initially derived from photo transect images using Coral Point Count with Excel extensions (CPCe; Kohler and Gill 2006) point sampling software, but the web-based image analysis platform CoralNet (Beijborn et al. 2012, Beijborn 2015) was used beginning with the 2019 images. Benthic percent cover is calculated at the sampling station level (= one transect) for data derived from benthic images. Power analyses support the methodological approach, whereby statistically robust cover estimates at broad-level benthic categories can be successfully derived using the level of sampling effort employed by the program.

The stated goals of the Guam Long-term Coral Reef Monitoring Program (GLTMP) are to determine the status and trends of selected coral reef ecosystem indicators to inform resource managers' decision-making process, provide managers with early notice of abnormal conditions, provide data to better understand the dynamic nature of the island's coastal ecosystems, allow resource agencies to meet certain legal and Congressional mandates, and measure progress towards performance goals. Benthic cover data collected by the GLTMP at several High Priority Reef Areas around Guam provide documentation of the structure and condition of the benthic communities at these sites; data from multiple visits can be used in time series analyses to determine trends in key ecosystem parameters.

Loaded by FGDC Metadata Uploader, batch 10200, 11-05-2017 15:30

The following FGDC sections are not currently supported in InPort, but were preserved and will be included in the FGDC export:

- Spatial Reference Information (FGDC:spref),

Benthic cover data have been generated from benthic photos captured since 2010 (with the noted exceptions below) in:

Tumon Bay in 2010 (10 permanent and 10 non-permanent stations), 2012 (12 permanent and 9 non-permanent stations), 2014 (12 permanent and 4 non-permanent stations), 2015 (12 permanent and 10 non-permanent stations), 2017 (12 permanent and 10 non-permanent stations), 2018 (12 permanent stations), 2019 (12 permanent and 10 non-permanent stations), 2020 (12 permanent and 10 non-permanent stations); 2021 (12 permanent and 10 non-permanent stations); and 2022 (12 permanent and 10 non-permanent stations)

East Agana Bay in 2010 (10 permanent and 10 non-permanent stations), 2012 (10 permanent stations), 2014 (10 permanent stations), 2015 (10 permanent stations), 2017 (10 permanent and 12 non-permanent stations), 2018 (10 permanent stations), 2019 (9 permanent and 6 non-permanent stations), 2020 (10 permanent and 10 non-permanent stations), 2021 (10 permanent and 10 non-permanent stations), and 2022 (10 permanent and 10 non-permanent stations)

Western Shoals in 2011 (11 permanent and 12 non-permanent stations)

Piti in 2012 (10 permanent and 10 non-permanent stations), 2014 (10 permanent and 2 non-permanent stations), 2017/18 (8 permanent and 10 non-permanent stations), 2020 (10 permanent and 10 non-permanent stations), and 2022 (10 permanent and 10 non-permanent stations)

Achang in 2014 (11 permanent and 3 non-permanent stations), 2018 (10 permanent and 10 non-permanent stations), and 2021 (10 permanent and 10 non-permanent stations)

Cocos-East in 2014 (3 permanent and 2 non-permanent stations), 2018 (10 permanent stations), and 2021 (10 permanent and 10 non-permanent stations)

Fouha Bay in 2015 (13 permanent stations), 2019 (13 permanent stations), and 2021 (13 permanent stations)

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, the University of Guam Marine Laboratory and the Guam Coastal Management Program 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. The University of Guam Marine Laboratory and the Guam Coastal Management Program make no warranty, expressed or implied, nor does the fact of distribution constitute such a warranty.

Tumon Bay

East Agana Bay

Western Shoals

Piti (Tepungan) Bay

Achang

Cocos-East

Fouha Bay

Not applicable

Not applicable

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

None

Please cite the Guam Coastal Management Program when using data collected prior to October 2013. Guam Coastal Management Program, Guam Bureau of Statistics and Plans, Government of Guam. Please cite the University of Guam Marine Laboratory when using data collected after (and including) October 2013.

Example citation:

For data collected from 2010 to 2012:

Guam Coastal Management Program (2019). Guam Long-term Coral Reef Monitoring Program Benthic Percent Cover Derived from Image Analysis since 2010. NOAA's National Center for Environmental Information, https://www.fisheries.noaa.gov/inport/item/47784.

For data collected after 2012:

University of Guam Marine Laboratory (2019). Guam Long-term Coral Reef Monitoring Program Benthic Percent Cover Derived from Image Analysis since 2010. NOAA's National Center for Environmental Information, https://www.fisheries.noaa.gov/inport/item/47784.

None

None

Benthic images are organized and edited on the desktop workstation of the Monitoring Program Coordinator, with backup copies of each image and associated data files stored on an external hard drive and with the GSuite cloud storage service. Between 2010 and 2018 images were analyzed using CPCe installations on the workstation or laptop computer of individual monitoring team members. CPCe output files generated by other team members were obtained by the coordinator and modified to conform to the directory structure of the coordinator's computer. Spreadsheets containing raw CPCe data from across multiple sites and years were generated by compiling .cpc files using CPCe and assembling the raw data within a single master spreadsheet. In 2021, benthic images obtained at the HPRAS beginning in 2019 were uploaded to CoralNet, a web-based image analysis platform. The CoralNet robot was initially trained using images and CPCe data from 2015-2018. Approximately 5% to 20% of annotations (depending on the site) suggested by the CoralNet robot for the 2019-2022 images exceeded the 80% confidence threshold set for the source, but a human observer still verified each annotation manually. Beginning with the 2023 images, a lower confidence threshold will be used and the CoralNet robot's annotations will be considered confirmed and will not be reviewed by a human observer. The use of a lower overall confidence threshold value is justified by the very high (>95%) accuracy obtained by the robot for targeted benthic classes.

Data sets derived from image analysis are highly reliable within the stipulated taxonomic context. Image analysts undergo a training period during which they analyze images from numerous sampling stations in the presence of the monitoring program coordinator. The total number of images analyzed during the training period varies according to the prior experience level of the trainee. The results of the CPCe image analyses (2010-2018) and the initial fully-manual CoralNet-based analyses (2019-2022) underwent extensive quality control procedures carried out by the benthic community specialist (currently the program coordinator). However, it should be understood that some benthic features, such as turf algae, crustose coralline algae, small coral colonies and other diminutive living or non-living features may be difficult to discern from each other in the benthic photo transect images. It should also be noted that the higher image resolution achieved by the more recently-used cameras allows for greater taxonomic resolution during image analysis, and likely allows for greater accuracy in the identification of benthic features, particularly for features, such as turf algae, crustose coralline algae, and small corals, that can be difficult to discern at lower image resolutions. Beginning with images obtained in 2023 the accuracy of the outputs of image analysis, which will involve a combination of human observer and CoralNet robot-alleviated annotations, will be evaluated using tools available on the CoralNet site (e.g., confusion matrix and confidence threshold sweep graph).

Effort is made to avoid inter-observer bias through consistent training of image analysis technicians by the Program Coordinator and by reference to an established protocol document. The use of the CoralNet platform for the semi-automated analysis of images obtained beginning in 2023 will further minimize inter-observer bias, but will introduce new biases.

Comparisons of data across HPRAs or across years must take into account differences in environmental and biological aspects of each site. Comparisons for a give HPRA across time should consider any changes to the site boundaries over time and the mix of paired and unpaired sampling stations. Please refer to the Extents section of this metadata record to see how the Tumon site boundaries shifted after 2010.

Comparisons for a given site across time could potentially be affected by changes in the benthic photo transect survey and image analysis methodologies. These changes are as follows:

Transect length: Benthic photo transect length for all sampling stations except the Western Shoals HPRA was 25 meters between 2010 and 2016 and, with the exception of the Fouha Bay HPRA, was extended to 30 m beginning in 2017.

Image spacing: Between 2010 and 2012 benthic photos were captured every 0.5 meters along both sides of the transect tape. Photos were taken every 1 meter beginning in 2014 in order to avoid overlapping images, and beginning in 2017 images were only taken along the left side of the transect.

Image analysis: Prior to 2014 a total of 25 points (random stratified using a 5 x 5 grid) were used to analyze each image in CPCe. For photos analyzed since (and including) 2014, a total of 16 points (random stratified using a 4 x 4 grid) are used.

Image quality: Image quality has improved over time with the availability of digital cameras capable of capturing higher resolution images and with more advanced functionality. A Canon PowerShot A720 IS digital camera (8 Megapixel sensor; 3264x2448 pixel image resolution) was used between 2010 and 2011, a Nikon Coolpix 36300 (16 Megapixel sensor; 4608x3456 pixel image resolution) was used for most sampling stations in 2012; and a Canon PowerShot ELPH 300 HS (12 Megapixel sensor; 4000x3000 pixel image resolution) was used for a limited number of sampling stations in 2012. The lower resolution PowerShot A720 was again used in 2014, and a Sony Cybershot RX-100 (20 Megapixel sensor; 5472x3648 pixel image resolution) has been used since 2015. The higher image resolution achieved by the more recent cameras allows for greater taxonomic resolution during image analysis, and likely allows for greater accuracy in the identification of benthic features, particularly for features, such as turf algae, crustose coralline algae, and small corals, that can be difficult to discern at lower image resolutions.

Camera frame/monopod: Initially, the camera/housing was mounted to a PVC frame/quadrat structure that allowed the consistent positioning of the camera directly above a 0.5 x 0.5 m quadrat. A 1.15 m-tall PVC monopod was used beginning in 2017. In order to minimize the presence of the monopod within the image frame the camera is mounted on a piece of PVC that extends 0.28 m horizontally at the top of the monopod. The field of view captured using the Sony RX100/monopod combination is approximately 50% larger than that obtained using the frame/quadrat.

Modification of benthic classification system: An effort was made to standardize benthic classes used in the analysis of all images to the extent possible, but care must still be taken in time series analyses of classes that had been split into two or more classes, or for which the definition had been modified at some point during the study period. Examples include the redefinition of TURF to refer to just microturf, the redefinition of TURS to refer to conspicuous turfing algae, the creation of the class CCRA to refer mainly to the Peyssonneliales and related encrusting coralline algae, the redefinition of FMAD to include only encrusting fleshy macroalgae (e.g., Lobophora). It is highly recommended that any users of these data reach out to the program coordinator if changes in the definition of these and some other classes may impact an analysis of these benthic cover data.

Only hard-bottom habitats are surveyed. Surveys are conducted annually to biennially for the Tumon and East Agana sites, but are conducted on a less frequent basis at other sites. In addition, while all permanent sampling stations are generally surveyed each sampling period, surveys are not always conducted at the full set of non-permanent stations for each site during each sampling period. Below is a summary of the number of permanent and non-permanent sampling stations surveyed each year for each site:

Tumon Bay: 2010 (10 perm, 10 non-perm); 2012 (12 perm, 9 non-perm); 2014 (12 perm, 4 non-perm); 2015 (12 perm, 10 non-perm); 2017 (12 perm, 10 non-perm); 2018 (12 perm, 0 non-perm); 2019 (12 perm, 10 non-perm); 2020 (12 perm, 10 non-perm); 2021 (12 perm,10 non-perm); 2022 (12 perm, 10 non-perm)

East Agana Bay: 2010 (10 perm, 10 non-perm); 2012 (10 perm, 0 non-perm); 2014 (10 perm, 0 non-perm); 2015 (10 perm, 0 non-perm); 2017 (10 perm, 12 non-perm); 2018 (10 perm, 0 non-perm); 2019 (9 perm, 6 non-perm); 2020 (10 perm, 10 non-perm); 2021 (10 perm, 10 non-perm); 2022 (10 perm, 10 non-perm)

Western Shoals: 2011 (11 perm, 12 non-perm)

Piti Bay: 2012 (10 perm, 10 non-perm); 2014 (10 perm, 2 non-perm); 2017-2018 (8 perm, 6 non-perm); 2020 (10 perm, 10 non-perm); 2022 (10 perm, 10 non-perm)

Achang: 2014 (11 perm, 3 non-perm); 2018 (9 perm, 6 non-perm); 2021 (10 perm, 10 non-perm)

Cocos-East: 2014 (3 perm, 4 non-perm); 2018 (10 perm); 2021 (10 perm, 10 non-perm)

Fouha Bay: 2015 (13 perm); 2019 (13 perm); 2021 (13 perm)

The analysis of images collected through 2022 is complete and the results have all undergone quality control. As of the publication of this metadata record images the analysis of images obtained in 2023 have not yet been completed.

Baseline benthic cover data derived from photo transects were explored in multivariate space using the statistical software package PRIMER and the PERMANOVA add-on. The exploration of the benthic cover data in multivariate space allowed for the visualization of the spatial structure of the data and the examination of the possible influence of environmental factors on this spatial structure. This preliminary data exploration also assisted in determining if the benthic community remained relatively homogeneous across the monitoring sites or strata as original delineated, and whether or not the site boundaries should be adjusted, subdivided into two or more strata, or if certain sampling stations should be abandoned. Univariate power analyses using the R software program were carried out separately on total coral cover for sampling stations from different strata or other sampling station groupings. Multivariate power analyses were carried out in order to examine the ability of the sampling regime to adequately capture the overall character of the benthic communities. The results of these analyses were presented in previous reports.

Due to time constraints, for most sampling stations only images collected along the left side of the transect were analyzed and underwent quality control. However, images collected along the right side of the transect may be analyzed at a later date to determine if the additional data points affect statistical power for target parameters. Beginning in 2017, images were only obtained along the left side of the transect.

The data set derived from the image analysis is only as good as the images themselves; poor images with low contrast, inappropriate white balance, or out of focus are discarded. Therefore, percent cover estimates for some sites may derive from differing number of points between sites. Although the analysis process is statistically robust to allow for differing n (points) between sites the users should be aware of this limitation.

Quality control for images obtained between 2010 and 2018 was carried out by the Monitoring Program Coordinator or the full-time technical support staff. The quality control process involved reviewing CPCe output files for all images for at least 25% of the total number of sampling stations for each monitoring site. If a significant number of misidentifications were encountered the original photo analyst was asked to undergo additional training and then re-analyze the images; alternatively, the coordinator or technical support staff would re-analyze all the images for the image sets in question. CPCe output files generated through an analysis of images carried out by the coordinator did not undergo a quality control check. After receiving extensive training, a single full-time technician analyzed images (using the CoralNet platform) for images obtained between 2019 and 2022. The technician regularly consulted with the coordinator throughout the analysis of these images in order to maintain a high degree of accuracy and the standardized implementation of the benthic classification system.

University of Guam Marine Lab resources and assets

The Benthic Photo Transect Survey methodology, employed by the Guam Long-term Coral Reef Monitoring Program since 2010