High-resolution species distribution modelling of the vulnerable marine ecosystem indicator species, Enallopsammia rostrata, on Walvis ridge, SE Atlantic

Knowledge on the distribution of Vulnerable Marine Ecosystems (VMEs) is critical to support their effective management. However, major knowledge gaps still exist, particularly in Areas Beyond National Jurisdiction (ABNJ). Species Distribution Models (SDMs) can help to fill these gaps, especially in historically under-sampled regions. The Walvis Ridge in the South-East Atlantic, for example, consists of extensive seamount complexes and hosts deep-sea fauna that constitute VMEs, but is among the most data-poor globally. Presence/background models predicting the occurrence of the cold-water coral Enallopsammia rostrata , the most frequently encountered VME indicator species in the area to date, were developed for Valdivia Bank and Ewing seamount (NE Walvis Ridge). We found that incorporating high-resolution hydrodynamic variables and applying modifications to compensate for class imbalance in presence/background models strengthened models and reduced uncertainty. Random Forest models generally outperformed MaxEnt models, but an ensemble model had the best overall performance. The ensemble model predicted that 4.45% (10 519.90 km 2 ) of the study area has suitable habitat for E. rostrata , driven by depth, slope, kinetic energy dissipation and downwelling intensity. We demonstrate that even when developed on smaller occurrence record datasets, SDMs can deliver useful predictions, given appropriate statistical compensations and the addition of high-resolution hydrodynamic variables. The maps produced represent a valuable addition to ongoing spatial planning in this ABNJ, and can guide future survey effort aiming to validate, strengthen and broaden SDMs to improve our understanding of the distribution of VMEs in the under-studied SE Atlantic.