Comparative species distribution model framework for marine conservation and its application to loggerhead turtles in the Mediterranean

Modeling species distributions in dynamic pelagic environments remains challenging, particularly for wide-ranging and highly mobile species when there is limited guidance on model performance. This limitation constrains the effective use of species distribution models (SDMs) in marine conservation, where robust and transferable predictions are essential for conservation planning. Using the loggerhead turtle (Caretta caretta) as a case study, we developed a benchmarking workflow to compare modeling strategies across spatial resolutions, environmental predictor variables, and algorithms. We evaluated five common modeling approaches-generalized linear models, generalized additive models, random forest, boosted regression trees, and MaxEnt-and used a 15-year dataset of visual surveys collected along standardized ferry-based transects across the Mediterranean Sea, with environmental predictor variables averaged across years to capture recurrent summer conditions. Model outputs were assessed for their ability to predict habitat suitability and space-use patterns in dynamic pelagic environments. At the same time, robustness and spatial transferability were evaluated using an independent, basin-scale aerial survey dataset. Our results revealed consistent patterns informing SDM selection for wide-ranging pelagic species. In offshore Mediterranean pelagic systems, increased spatial resolution improved the ecological interpretability of predicted habitat patterns, whereas an extended set of dynamic oceanographic predictors enhanced the characterization of habitat associations. Independent validation was critical for selecting robust and transferable modeling algorithms, with MaxEnt and generalized additive models providing the most reliable performance for suitability and encounter rate (ER) modeling, respectively. Combining presence-only and encounter rate models helped distinguish habitat suitability from observation intensity. Overall, this study developed an empirically tested workflow to support robust, consistent, and policy-relevant SDM applications for mobile marine species in offshore, data-limited systems. By clarifying trade-offs among modeling approaches, this framework enhanced the application of SDMs to inform spatial planning, conservation prioritization, and species distribution and habitat-use assessments within regional and international conservation frameworks.