Li, Wenxiong and Ying, Yiping and Zhang, Jichang and Zhao, Yunxia and Zhu, Jiancheng and Fan, Gangzhou and Mu, Xiuxia and Wang, Xinliang (2025) The distribution modeling and analysis of Antarctic krill: impacts of algorithm and spatial resolution. Advances in Polar Science, 36 (4). pp. 373-391.
|
PDF
- Published Version
Download (1MB) | Preview |
Abstract
Antarctic krill (Euphausia superba), widely distributes around Antarctica, is a key species supporting the biodiversity of the Southern Ocean ecosystem. The Commission for the Conservation of Antarctic Marine Living Resources (CCAMLR) has thus managed the krill fishery according to a precautionary way. Currently, CCAMLR is making effort to develop a refined krill fishery management approach based on more solid science, which requires accurate predictions of krill distribution. To address this need, this study investigated the effects of algorithm and spatial resolution on the performance of Antarctic krill distribution modelling. We integrated acoustic data from 4 surveys conducted in the waters adjacent to the Antarctic Peninsula with 11 environmental variables characterizing krill prey conditions, water mass properties, and seafloor topography. These data were processed at 4 spatial resolutions (5, 10, 15, and 20 km) to fit distribution models using 4 algorithms: Random Forests (RF), Generalized Additive Models (GAM), Extreme Gradient Boosting (XGBoost), and Artificial Neural Networks (ANN). Model performance was assessed and compared in terms of goodness-of-fit and predictive accuracy. The results showed that RF achieved the highest predictive performance at most resolutions, whereas GAM performed best at the coarsest resolution (20 km). XGBoost closely following RF in accuracy and demonstrated robustness as evidenced by the highly consistent partial dependence curves across resolutions. In contrast, ANN exhibited limitations with smaller sample sizes, resulting in comparatively poorer predictive performance. The analysis revealed a trade-off whereby reducing spatial resolution improved model fit and mitigated zero-inflation at the expense of fine-scale information and overall predictive accuracy. Ensemble models, integrating RF, GAM, and XGBoost, are proposed as potential balanced solutions to improve predictive stability, offering a more robust scientific basis for the refinement of krill management.
| Item Type: | Article |
|---|---|
| Related URLs: | |
| Keywords: | Antarctic krill, species distribution model, algorithm selection, spatial resolution, machine learning |
| Subjects: | Natural Environment > Fauna Peoples, Cultures and Societies > Media |
| Organizations: | Advances in Polar Science (APS) |
| Date Deposited: | 24 Apr 2026 08:55 |
| URI: | https://library.arcticportal.org/id/eprint/2954 |
Actions (login required)
 |
View Item |