Coral reef benthic imagery and derived cover estimates from diver, drop camera and ReefScan surveys in the eastern Torres Strait (2021-2023)

Updated: 2026-08-13

This dataset supports the co-development of culturally informed and scientifically robust Indigenous-led coral reef monitoring in Kemer Kemer Meriam Nation, eastern Torres Strait, Australia. The data were collected to establish baseline information on coral reef habitat condition, compare the effectiveness of emerging image-based monitoring technologies, and identify monitoring approaches that align with Traditional Owner stewardship objectives, operational requirements, and long-term sea Country management needs. The research contributes to improved understanding of reef condition in a region recognised for its ecological, cultural, and potential climate-refugia significance. Data were collected from coral reefs surrounding Erub (Darnley Island), Ugar (Stephen Island), and Mer (Murray Island) in the eastern Torres Strait, approximately 180 km north of Cape York, Queensland, Australia. Surveys were undertaken at shallow reef-slope habitats (approximately 3-6 m depth) across six monitoring sites in 2021 and nine monitoring sites in 2023. Field campaigns were conducted between 28 May and 3 June 2021, 12-21 March 2023, and 26 November-7 December 2023. Site selection was undertaken collaboratively between Traditional Owners and researchers to ensure both scientific robustness and suitability for long-term community-led monitoring. The dataset consists of digital coral reef imagery collected using three complementary survey methods: SCUBA-based diver photo-transects, vessel-deployed drop cameras, and the vessel-mounted ReefScan imaging system. Diver surveys followed standardised AIMS photo-transect protocols using underwater digital cameras positioned approximately 50 cm above the substrate along 50 m transects. Drop-camera surveys used downward-facing cameras mounted on a tripod frame and deployed from small vessels along diver survey tracks. ReefScan surveys employed a transom-mounted camera system operated from vessels to collect georeferenced imagery along approximately 2 km reef-slope transects. Across all surveys, 166,454 images were acquired, including 7,442 diver images, 7,998 drop-camera images, and 151,014 ReefScan images. Image data were analysed using ReefCloud, a cloud-based artificial intelligence platform that applies deep-learning image classification models to estimate benthic community composition. Separate machine-learning models were developed and validated for each survey method using expert-annotated training data. Images were classified into ecologically relevant benthic groups including hard corals, soft corals, turf algae, sand, and rare taxa. Quality-control procedures included manual expert verification of training annotations, image-quality filtering, hierarchical spatial clustering to define comparable survey areas, and harmonisation of taxonomic classifications across methods. The data are interpreted through Bayesian mixed-effects statistical models that estimate benthic cover, quantify uncertainty, and evaluate differences among survey methods, monitoring sites, and islands. Diver photo-transects were used as a reference method for assessing agreement among monitoring approaches. ReefScan data were subsequently used to generate a regional baseline assessment of benthic community composition across the eastern Torres Strait. In addition to ecological analyses, the dataset incorporates Traditional Owner assessments of monitoring suitability, allowing interpretation of monitoring performance in relation to scientific accuracy, operational safety, spatial coverage, ease of use, and Indigenous governance objectives. The resulting data products provide a contemporary baseline for future assessments of reef condition and support the detection of changes associated with coral bleaching, crown-of-thorns starfish outbreaks, and other environmental pressures.