Body Length, Tow Conditions, and Guide-Net Angle as Joint Predictive Contributors to Catch Loss in a Marine Mammal Bycatch Reduction Device: An Interpretable Machine Learning Analysis
Fishes · 16.07.2026 · 11 (7) · 416
DOI: 10.3390/fishes11070416 ↗Kurz gefasst
Fachartikel in der Zeitschrift Fishes, online erschienen am 16.07.2026. Verfasst von Kyu-Suk Choi und 2 weiteren Autor:innen. Die Arbeit nennt 31 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: National Institute of Fisheries Science (NIFS) of Korea.
Abstract (Originalsprache)
Guide-net marine mammal bycatch reduction devices (BRDs) are intended to reduce protected species interactions in trawl fisheries, but their use may also cause catch loss of commercial fish and cephalopods. We restructured June and September 2024 sea trial records into a final six-taxon BRD/COD modeling dataset containing 5641 individual records from 24 unique haul clusters. Logistic GLM, random forest, and XGBoost models were evaluated using haul-grouped cross-validation, and random forest SHAP and permutation analyses were used as exploratory variable contribution diagnostics. Model discrimination was modest and split-sensitive, but Brier score and accuracy were interpreted cautiously because the catch loss prevalence was 0.163 and trivial baselines were strong. Body length and headline height were the leading non-species contributors, while nominal guide-net angle remained a relevant design variable but ranked below several biological, tow condition, and realized gear geometry variables. Additional haul-clustered GEE sensitivity analyses treated nominal guide-net angle as a two-level categorical factor (30°/45°). Across four GEE specifications, the 45° angle term was consistently positive, but confidence intervals were wide, p-values were not significant, and the full species-adjusted model showed suspected quasi-separation in the hairtail term. These results indicate that nominal angle contribution cannot be cleanly isolated from species composition and realized gear geometry in this field dataset. We therefore interpret guide-net angle as an exploratory predictive contributor rather than as a causal factor of catch loss.
Förderung
- National Institute of Fisheries Science (NIFS) of Korea
Zitieren
APA
Choi, K., Jung, J., & Kim, H. (2026). Body Length, Tow Conditions, and Guide-Net Angle as Joint Predictive Contributors to Catch Loss in a Marine Mammal Bycatch Reduction Device: An Interpretable Machine Learning Analysis. Fishes, 11(7), 416. https://doi.org/10.3390/fishes11070416
Vancouver
Choi K, Jung J, Kim H. Body Length, Tow Conditions, and Guide-Net Angle as Joint Predictive Contributors to Catch Loss in a Marine Mammal Bycatch Reduction Device: An Interpretable Machine Learning Analysis. Fishes. 2026;11(7):416. doi:10.3390/fishes11070416
BibTeX
@article{Choi2026,
author = {Choi, Kyu-Suk and Jung, Jung-Mo and Kim, Hyun-Young},
title = {Body Length, Tow Conditions, and Guide-Net Angle as Joint Predictive Contributors to Catch Loss in a Marine Mammal Bycatch Reduction Device: An Interpretable Machine Learning Analysis},
journal = {Fishes},
year = {2026},
volume = {11},
number = {7},
pages = {416},
doi = {10.3390/fishes11070416}
}
EUVIMED