Bayesian clustering of spatially distributed compositional data with application to the Great Barrier Reef.

Area of Science:

• Marine Ecology
• Statistical Ecology
• Conservation Biology

Background:

• Community composition, reflecting species commonness and rarity, is crucial for ecological surveys.
• Benthic group abundances on the Great Barrier Reef (GBR) provide insights into seafloor ecosystems.

Purpose of the Study:

• To develop a statistical model for identifying clusters of reefs with similar benthic compositions.
• To examine how reef composition clusters changed during a period of extreme environmental disturbances on the GBR.

Main Methods:

• A Dirichlet mixture model incorporating geographical location and a Potts distribution was used to cluster reef sites.
• An approximate exchange algorithm was employed to address Bayesian inference challenges with the non-normalized Potts distribution.
• Analysis of benthic group abundances from 2012 to 2017 on the GBR.

Main Results:

• In 2012, four distinct clusters of GBR reefs were identified, characterized by specific habitat patterns (soft coral abundance, sand dominance, balanced species distribution, algae/hard coral dominance).
• Post-disturbance surveys indicated a reduction in the number of clusters.
• A simplification of reef composition at the regional scale was observed following environmental disturbances.

Conclusions:

• The study successfully modeled spatial patterns of benthic community composition on the GBR.
• Extreme environmental disturbances significantly impacted reef structure, leading to a less diverse and simplified regional composition.
• The developed statistical approach offers a robust method for analyzing ecological survey data and assessing environmental impacts.