Predicting coral trends and attributing drivers of change from local to global scales

Modern biodiversity monitoring programs are designed to assess abundance trends of keystone taxa and deliver scientific insights to inform decision-making and policy development. An important consideration when using these evidence-based frameworks is the quantification of uncertainty from trends, which determines the robustness of data-driven methods in detecting and attributing changes across habitats and regions. In coral reefs, sparse and fragmented monitoring programs challenge the assessment of long-term reef habitat changes, despite the need for actionable insights to reduce the loss of coral cover. We introduce a new predictive modelling framework, which combines machine learning for the extraction of ecological data with a statistical model to predict trends in hard coral cover and propagate uncertainty across multiple spatial scales. The model estimates the spatio-temporal variability of coral cover at monitoring locations and integrates information on marine heatwaves and cyclones to predict coral cover across entire marine ecoregions, thereby filling the spatial and temporal observational gaps inherent in coral reef monitoring programs. It also quantifies the effects of known regional drivers on coral cover loss and allows the exploration on how specific disturbances influence coral cover spatially across regions. We demonstrate the framework's capability using case studies from the northern Great Barrier Reef and in simulation experiments. Together, these illustrate the importance of incorporating the spatial dimension to capture the variability in coral cover and attribute drivers of coral cover change with greater confidence. This modelling framework is designed for integration into the ReefCloud platform, where it can automatically combine data from monitoring programs worldwide and support evidence-based decision-making for the management and conservation of coral reefs from local to global scales.