AniDMS: Estimating dimethyl sulphide (DMS) across North Atlantic over 20 years with machine learning

Existing dimethyl sulphide (DMS) products are too coarse in space and time (≥25 km, monthly) to support movement-scale analysis of seabird olfactory navigation. This leads to a geospatial mismatch between environmental representations and individual movement trajectories. We developed a machine-learning framework that combines multi-source data harmonisation, spatio-temporal interpolation, and stacking-based ensemble prediction to reconstruct daily DMS fields at 4km resolution across the North Atlantic. Six base learners were evaluated and integrated using an ElasticNet stacking design. The final model achieved strong performance (test R² = 0.88; RMSE = 0.859 µmol m⁻³), outperforming previously reported basin-scale approaches. The resulting product captures seasonal cycles, persistent hotspots, and fine-scale gradients. We further developed AniDMS, an open-source Python tool for automated retrieval of gridded DMS data and reproducible annotation of trajectories with DMS. A Manx shearwater case study demonstrates the integration of a dynamic DMS surface with movement data. This work provides a transferable geospatial workflow for reconstructing high-resolution environmental fields and linking them to tracking data.