Dr Stanislas DUBOIS1, Dr Aurélien BOYE1, Dr Martin MARZLOFF1, Dr Laura BUSH2, Céline CORDIER1, Dr Amelia CURD1, Dr Andrew J. DAVIES3, Dr Louise B. FIRTH4, Dr Fernando LIMA5, Dr Claudia MENEGHESSO6, Dr Rui SEABRA5, Mickael VASQUEZ1
1Ifremer Centre Bretagne, Plouzané, France, 2Fugro GB Marine Limited, Edinburgh, United Kingdom, 3Department of Biological Sciences, University of Rhode Island, Kingston, USA, 4School of Biological and Marine Sciences, University of Plymouth, Plymouth, United Kingdom, 5CIBIO/InBIO, Centro de Investigação em Biodiversidade e Recursos Geneticos, Universidade do Porto, Vairão, Portugal, 6Departamento de Biologia, Faculdade de Ciências da Universidade do Porto, Porto, Portugal
Marine ecosystems can exhibit abrupt regime shifts, in which even small changes in environmental conditions lead to fundamentally different and persistent ecosystem states. Regime shifts can profoundly alter the delivery of ecosystem services and may be hard to reverse, prompting the need to better predict their occurrence and dynamics to effectively inform management and conservation. Here, we applied machine-learning techniques to data from a cost-effective and rapid monitoring protocol targeting intertidal Sabellaria alveolata reefs, a European-widespread gregarious tubiculous species building extensive structures. Based biannual surveys monitoring ecological changes across 12 sites across Europe, we identify 5 stable and 1 transient reef states. These states can emerge in similar environments, which qualifies them as alternative stable states. Their predicted probability of occurrence changes non-linearly along environmental gradients, and we could identify several key thresholds in abiotic conditions beyond which catastrophic transitions between alternative regimes are likely. Prediction models can reliably discriminate between two sets of alternative states, one dominated by Sabellaria bioconstructions and one dominated by other rocky-reef species. Spatially-explicit model predictions allowed us to characterise expected dominant reefs states in data-poor regions, identify zones that are likely to host alternative stable states at a given time, as well as areas prone to environmentally-driven regime shifts. Machine learning approaches successfully predict the non-linear and context-dependent responses of reef states to environmental changes and our study promotes the novel role of iterative ecological forecasts to inform effective reef conservation strategies that accounts for shifting environmental baselines.
Biography:
Stanislas Dubois is a researcher in biological oceanography at IFREMER, France. He specialized in biodiversity patterns of marine benthic species and more particularly in ecosystem engineer species. Through the use of complementary approaches (in situ sampling, spatial imagery, mesocosm experiments), he studies the physical processes and biological mechanisms that spatially structure benthic marine communities and their changes over time.
