Benthic Community Imagery from EcoRRAP Reef Monitoring 2021-2024

EcoRRAP photogrammetric monitoring produced benthic community composition data from 352 plots in 2021, 333 in 2022, 325 in 2023, and 191 in 2024. Images were analyzed using a ReefCloud ML model trained to 80% accuracy on a custom label set, with twelve 1000-pixel-bounded images extracted per plot. The dataset includes machine-generated annotations for 2024 and human-annotated training data for prior years.

Use Cases

• Train computer vision models on the EcoRRAP Community Composition Label Set to classify benthic organisms from reef imagery.
• Analyze temporal changes in benthic community composition across 2021-2024 using plot-level image annotations.
• Validate the performance of automated image analysis models against human-annotated points from 2021-2023.
• Assess spatial stratification of benthic communities using images filtered by position within 3D photogrammetric models.

Strengths

• Data spans four consecutive years (2021-2024) with imagery from over 1,200 total monitoring plots.
• Includes a custom EcoRRAP Community Composition Label Set for consistent taxonomic annotation.
• ML model was iteratively trained with a defined 80% accuracy target and underwent a full quality check in January 2026.
• Original high-resolution imagery (8256x5504 pixels) was processed with a 1000-pixel bounding box to reduce edge distortion.

Limitations

• 2024 data is limited to reefs in the Central and Torres Strait clusters and contains only machine-generated annotations, with no human quality check.
• Sample size decreases year-over-year, with 191 plots in 2024 compared to 352 in 2021, and missing plots are unavailable due to logistical constraints.
• The dataset's primary format is imagery (PNG) with associated annotations, requiring specialized computer vision tools for quantitative analysis.

Provenance

Source

Australian Ocean Data Network via data_gov_au.

Collection Method

Generated from diver-collected imagery analyzed via photogrammetry and a ReefCloud machine learning model, using a defined annotation protocol (SOP #14).

Time Range

2021 to 2024.

Freshness

Data was last updated on March 22, 2026.

Geography

EcoRRAP monitoring sites, specifically the Central Cluster and Torres Strait Cluster for 2024 data.