Using Machine Learning Methods to Estimate Sea Surface DMS Concentration for Studying Seabird Movements

Liu, Meixuan; Benitez-Paez, Fernando; Padge, Oliver; Kishkinev, Dmitry; Demšar, Urška

Abstract

Dimethylsulphide (DMS) has been shown to play a vital role in foraging behaviour of seabirds and suggested as a key component of olfactory navigation, yet its spatial distribution remains poorly understood due to limited point-based sampling. This study evaluates five machine learning algorithms and three ensemble models to estimate sea surface DMS concentrations in the North Atlantic using remote-sensing sources. XGBoost and an ensemble model incorporating ridge regression indicate high accuracy and outperform other methods. Also, models reflect that DMS levels are strongly influenced by nitrate and chlorophyll-a, aligning with their chemical properties and increasing the reliability of the model. The proposed approach aims to provide accurate, high-resolution DMS concentration estimates to support future research on avian olfactory navigation.

Citation

Liu, M., Benitez-Paez, F., Padge, O., Kishkinev, D., & Demšar, U. (2025, April). Using Machine Learning Methods to Estimate Sea Surface DMS Concentration for Studying Seabird Movements. Presented at 33rd GISRUK Conference 2025, University of Bristol, Bristol, UK

• Presentation Conference Type: Conference Paper (published)
• Conference Name: 33rd GISRUK Conference 2025
• Start Date: Apr 23, 2025
• End Date: Apr 25, 2025
• Acceptance Date: Apr 1, 2025
• Online Publication Date: Apr 17, 2025
• Publication Date: Apr 17, 2025
• Deposit Date: Apr 29, 2025
• Publicly Available Date: Apr 29, 2025
• Publisher: Zenodo
• Peer Reviewed: Peer Reviewed
• DOI: https://doi.org/10.5281/zenodo.15237116
• Keywords: Dimethylsulphide (DMS), Remote Sensing, Machine Learning, Animal Movement
• Public URL: https://keele-repository.worktribe.com/output/1202260
• Publisher URL: https://zenodo.org/records/15237117
• External URL: https://zenodo.org/records/15237117
• Other Repo URL: https://zenodo.org/records/15237117