Machine-learned dimethyl sulphide (DMS) for the North Atlantic (2002–2024) to support movement studies

Dimethyl sulphide (DMS) serves as a key olfactory cue for seabird navigation, yet existing DMS products operate at coarse spatiotemporal resolutions (≥ 25 km, monthly) mismatched to the scales of individual movement decisions. We ask (I) whether machine learning can produce biologically relevant, high-resolution DMS estimates across the North Atlantic, and (ii) whether such estimates can be readily integrated with animal tracking data to support ecological interpretation of animal trajectories. Using North Atlantic in-situ DMS observations (2002–2024) and five satellite-data-based environmental predictors (chlorophyll, mixed layer depth, nitrate, sea-surface temperature, and photosynthetically available radiation), we developed a machine-learning-based ensemble model for DMS prediction that achieved strong accuracy (test R² = 0.88; RMSE = 0.859 µmol m⁻³), exceeding previously reported performance for basin-scale DMS mapping. To identify the key drivers of the model predictions, we conducted SHAP (SHapley Additive exPlanations) analysis, which revealed that mixed layer depth, nitrate concentration, and chlorophyll were the dominant controlling factors, aligning with established understanding of DMS biogeochemistry. We then produced a spatially continuous, daily 4 km DMS dataset for the North Atlantic domain (0–60° N, 80° W–15° E), revealing seasonal cycles, persistent hotspots, and fine-scale gradients not captured by coarse climatologies. Finally, we develop AniDMS, an open-source Python package that automates trajectory annotation with gridded DMS and associated covariates, demonstrated with a Manx shearwater case study. Together, the dataset and the tool enable scalable, hypothesis-driven tests of olfactory navigation of seabirds and provide a transferable framework for integrating high-resolution environmental context into movement ecology.