A change is gonna come: predicting degradation and recovery dynamics in intertidal biogenic reefs using machine learning and spatial early warning signalsResistance, Resilience and Phase Shifts

Aurélien Boyé¹, Martin P. Marzloff¹, Céline Cordier¹, Stanislas F. Dubois¹

¹Ifremer, France

Marine ecosystems are prone to abrupt regime shifts, where small environmental changes can lead to fundamentally different and persistent states. These shifts can severely impact ecosystem functions and are often difficult to reverse, emphasizing the need for predictive tools to guide early and effective conservation and management interventions. In this study, we applied machine-learning techniques to biannual monitoring data from Sabellaria alveolata reefs, a gregarious tubiculous species that constructs dynamic intertidal biogenic reefs across European coastlines. These reefs, ranging from thin veneer-like encrustations to large hummock-like structures, commonly settle and develop over rocky substrates, exhibiting natural cycles of construction and degradation that remain challenging to anticipate and manage.

Using data from 12 monitoring sites across Europe, we identified five stable and one transient reef state. We then combined spatial early warning signals (e.g., changes in spatial heterogeneity) with a machine-learning model to predict short-term transition probabilities between reef states. Independent validation against recent monitoring data suggests that the model satisfactorily estimates probabilities of ecological transitions, either (1) reef decline to non-reef states, (2) reef recovery, or (3) state persistence over consecutive six-monthly surveys. Model performance is particularly reliable in regions where longer-term monitoring data is available, which highlights the value of standardized long-term ecological surveys.

These transition probabilities offer actionable insights into reef degradation risks and have been incorporated into management frameworks, such as the Marine Strategy Framework Directive (MSFD). By integrating machine learning with spatial early warning signals, this study provides an innovative approach for anticipating degradation and recovery dynamics in reef ecosystems. Our short-term predictive framework supports iterative ecological forecasting, which can help guide conservation strategies and proactive management interventions that account for both current reef conditions as well as potential future risks of change.