A Machine Learning Species Distribution Model for the Critically Endangered East Pacific Leatherback Turtle Dermochelys Coriacea

The Eastern Pacific population of leatherback turtles Dermochelys coriacea is Critically Endangered, with incidental capture in coastal and pelagic fisheries as one of the major causes. Given the population’s broad geographic range, status, and extensive overlap with fisheries throughout the region, identifying areas of high importance is essential for effective conservation and management. In this study, we created a machine-learning species distribution model trained with remotely sensed environmental data and fishery-dependent leatherback presence (n = 1088) and absence data (>500000 fishing sets with no turtle observations) from industrial and small-scale fisheries that operated in the eastern Pacific Ocean between 1995 and 2020. The data were obtained through a participatory collaboration between the Inter-American Convention for the Protection and Conservation of Sea Turtles and the Inter-American Tropical Tuna Commission as well as non-governmental organizations to support the quantification of leatherback vulnerability to fisheries bycatch. A daily process was applied to predict the probability of leatherback occurrence as a function of dynamic and static environmental covariates. Coastal areas throughout the region were highlighted as important habitats, particularly highly productive feeding areas over the continental shelf of Ecuador, Peru, and offshore from Chile, and breeding areas off Mexico and Central America. Our model served as the basis to quantify leatherback vulnerability to fisheries bycatch and the potential efficacy of conservation and management measures (Griffiths & Wallace et al. 2024; Endang Species Res 53:295-326). In addition, this approach can provide a modeling framework for other data-limited vulnerable populations and species.