Predicting coral cover trends from local to broad spatial scales

Julie Vercelloni, Murray Logan, Andrew Zammit-Mangion, Matthew Sainsbury-Dale, Britta Schaffelke, Kerrie Mengersen, Manuel Gonzalez Rivero

Modern biodiversity monitoring programs are designed to assess trends in abundance and distribution of keystone taxa and deliver scientific insights to inform decision-making and policy development. An important consideration when applying this information is the quantification of uncertainty, which determines the robustness of detecting changes across habitats and regions. In coral reefs, sparse and fragmented monitoring datasets hinder the assessment of reef habitat changes. We introduce a new spatio-temporal prediction framework for estimating trends in hard coral cover and associated uncertainty at a local scale (5 km² spatial units). The model accounts for spatial and temporal dependencies in coral cover through a latent process formulation, where correlation is structured explicitly in space and time, and includes environmental variables describing exposure to heat stress and tropical cyclones. We use a weighted spatial aggregation approach to predict trends at the regional scale, with the spatial extent of a region varying according to the geographic context. The same approach is applied to estimate trends at broader spatial scales by combining outputs from multiple models through the ReefCloud platform, ensuring that predictions reflect the spatial distribution of reef-building corals and that uncertainty is appropriately propagated across spatial scales. The model also quantifies effects of heat stress and tropical cyclones and characterizes their associations with coral cover change across gradients of disturbance intensity. We demonstrate the value of the framework using three use cases: the central Great Barrier Reef in Australia, American Samoa and simulation experiments. Together, these applications highlight the potential of our integrated approach as a widely applicable method for predicting coral cover trends at various spatial scales. We also discuss the substantial uncertainty associated with the framework due to the limitations of available datasets and suggest approaches to improve the robustness of trend detection for coral reefs and their attribution to environmental disturbances.