Modelling the distribution of sea turtles in the Western Indian Ocean based on bycatch data from the French pelagic longline and purse seine fisheries

Species Distribution Models (SDMs) are valuable predictive tools to anticipate bycatch risk in fisheries. Bycatch of sea turtles, which are of conservation concern worldwide, could negatively affect populations through direct mortality or decreased post-release fitness. With a better understanding of the environmental variables driving their distribution, one could provide successful bycatch mitigation strategies. However, this remains an important knowledge gap for sea turtles in the Western Indian Ocean. To address this, we used two modelling approaches, namely logistic regression and Random Forest, to identify and quantify the importance of 15 candidate environmental predictors for loggerhead (TTL), olive ridley (LKV), and green (TUG) turtles. Using on-board observer data from the French pelagic longline and purse seine fisheries, we show that sea surface height and the Dipole Mode Index could be important predictors of bycatch events for the three turtle species. Our results should prove useful to select appropriate environmental variables depending on the focal species to fit SDMs from bycatch data. Nevertheless, the modelling approaches used here have limitations that warrant consideration. We discuss those and provide recommendations for further improvement.