Interpretable fish abundance index prediction in tuna longline fisheries: A LightGBM-SHAP case study in the tropical Atlantic Ocean

IF 2.3 2区 农林科学 Q2 FISHERIES
Linhui Wang , Liming Song , Hengshou Sui , Bin Li
{"title":"Interpretable fish abundance index prediction in tuna longline fisheries: A LightGBM-SHAP case study in the tropical Atlantic Ocean","authors":"Linhui Wang ,&nbsp;Liming Song ,&nbsp;Hengshou Sui ,&nbsp;Bin Li","doi":"10.1016/j.fishres.2025.107468","DOIUrl":null,"url":null,"abstract":"<div><div>Accurately predicting fish abundance index is crucial for sustainable fisheries management. This study focuses on three highly migratory fish species: bigeye tuna (<em>Thunnus obesus</em>), yellowfin tuna (<em>Thunnus albacares</em>), and swordfish (<em>Xiphias gladius</em>) in the tropical Atlantic Ocean (TAO). Utilizing tuna longline logbook data from 2016 to 2019 and various environmental datasets, we employed four feature selection methods: no processing, correlation analysis with multicollinearity diagnosis, traditional Principal Component Analysis (PCA), and stratified PCA. Seven predictive models for abundance indices were rigorously compared to identify optimal modeling frameworks and feature engineering methodologies. An interpretable LightGBM-SHAP model was subsequently developed to predict CPUE while quantifying the relative contributions of key environmental drivers. The framework’s spatial applicability was verified using the KDE tool, Moran’s I index, and two correlation analyses. Results demonstrated that utilization of raw environmental variables without dimensionality reduction yielded superior predictive performance (R<sup>2</sup>&gt;0.84 across all species), underscoring the necessity of context-appropriate feature selection. Spatial validation confirmed strong concordance between SHAP-derived predictions and observed CPUE distributions. Critical species-specific environmental determinants were identified: (1) the most influential factors were month, longitude, and latitude for bigeye tuna; (2) latitude, month, and D250 were the dominant factors for yellowfin tuna; (3) latitude, month, and D450 were key factors for swordfish. This study provides a comprehensive framework for predicting fish abundance index and interpreting the underlying environmental factors, thereby enhancing the interpretability of machine learning models in fisheries forecasting. The findings offer valuable insights for fisheries managers to identify potential fishing zones, adjust management strategies, and promote the sustainable utilization of fisheries resources in the TAO.</div></div>","PeriodicalId":50443,"journal":{"name":"Fisheries Research","volume":"288 ","pages":"Article 107468"},"PeriodicalIF":2.3000,"publicationDate":"2025-07-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Fisheries Research","FirstCategoryId":"97","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S016578362500205X","RegionNum":2,"RegionCategory":"农林科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q2","JCRName":"FISHERIES","Score":null,"Total":0}
引用次数: 0

Abstract

Accurately predicting fish abundance index is crucial for sustainable fisheries management. This study focuses on three highly migratory fish species: bigeye tuna (Thunnus obesus), yellowfin tuna (Thunnus albacares), and swordfish (Xiphias gladius) in the tropical Atlantic Ocean (TAO). Utilizing tuna longline logbook data from 2016 to 2019 and various environmental datasets, we employed four feature selection methods: no processing, correlation analysis with multicollinearity diagnosis, traditional Principal Component Analysis (PCA), and stratified PCA. Seven predictive models for abundance indices were rigorously compared to identify optimal modeling frameworks and feature engineering methodologies. An interpretable LightGBM-SHAP model was subsequently developed to predict CPUE while quantifying the relative contributions of key environmental drivers. The framework’s spatial applicability was verified using the KDE tool, Moran’s I index, and two correlation analyses. Results demonstrated that utilization of raw environmental variables without dimensionality reduction yielded superior predictive performance (R2>0.84 across all species), underscoring the necessity of context-appropriate feature selection. Spatial validation confirmed strong concordance between SHAP-derived predictions and observed CPUE distributions. Critical species-specific environmental determinants were identified: (1) the most influential factors were month, longitude, and latitude for bigeye tuna; (2) latitude, month, and D250 were the dominant factors for yellowfin tuna; (3) latitude, month, and D450 were key factors for swordfish. This study provides a comprehensive framework for predicting fish abundance index and interpreting the underlying environmental factors, thereby enhancing the interpretability of machine learning models in fisheries forecasting. The findings offer valuable insights for fisheries managers to identify potential fishing zones, adjust management strategies, and promote the sustainable utilization of fisheries resources in the TAO.
金枪鱼延绳钓渔业中可解释的鱼类丰度指数预测:热带大西洋LightGBM-SHAP案例研究
准确预测鱼类丰度指数对渔业可持续管理至关重要。本研究以热带大西洋(TAO)中的三种高度洄游鱼类:大眼金枪鱼(Thunnus obesus)、黄鳍金枪鱼(Thunnus albacares)和剑鱼(Xiphias gladius)为研究对象。利用2016 - 2019年金枪鱼延绳钓日志数据和各种环境数据集,采用无处理、多重共线性诊断相关分析、传统主成分分析(PCA)和分层主成分分析(PCA)四种特征选择方法。对7种丰度指数预测模型进行了严格比较,以确定最佳建模框架和特征工程方法。随后开发了一个可解释的LightGBM-SHAP模型来预测CPUE,同时量化关键环境驱动因素的相对贡献。使用KDE工具、Moran’s I指数和两个相关分析验证了该框架的空间适用性。结果表明,未降维的原始环境变量的利用产生了更好的预测性能(在所有物种中R2>;0.84),强调了与环境相适应的特征选择的必要性。空间验证证实了shap推导的预测与观测到的CPUE分布之间的强一致性。结果表明:(1)对大眼金枪鱼影响最大的环境因子为月份、经度和纬度;(2)纬度、月份和D250是黄鳍金枪鱼的优势因子;(3)纬度、月份和D450是剑鱼生长的关键因子。本研究为预测鱼类丰度指数和解释潜在的环境因素提供了一个综合框架,从而提高了机器学习模型在渔业预测中的可解释性。研究结果为渔业管理者识别潜在渔区、调整管理策略和促进渔业资源的可持续利用提供了有价值的见解。
本文章由计算机程序翻译,如有差异,请以英文原文为准。
求助全文
约1分钟内获得全文 求助全文
来源期刊
Fisheries Research
Fisheries Research 农林科学-渔业
CiteScore
4.50
自引率
16.70%
发文量
294
审稿时长
15 weeks
期刊介绍: This journal provides an international forum for the publication of papers in the areas of fisheries science, fishing technology, fisheries management and relevant socio-economics. The scope covers fisheries in salt, brackish and freshwater systems, and all aspects of associated ecology, environmental aspects of fisheries, and economics. Both theoretical and practical papers are acceptable, including laboratory and field experimental studies relevant to fisheries. Papers on the conservation of exploitable living resources are welcome. Review and Viewpoint articles are also published. As the specified areas inevitably impinge on and interrelate with each other, the approach of the journal is multidisciplinary, and authors are encouraged to emphasise the relevance of their own work to that of other disciplines. The journal is intended for fisheries scientists, biological oceanographers, gear technologists, economists, managers, administrators, policy makers and legislators.
×
引用
GB/T 7714-2015
复制
MLA
复制
APA
复制
导出至
BibTeX EndNote RefMan NoteFirst NoteExpress
×
提示
您的信息不完整,为了账户安全,请先补充。
现在去补充
×
提示
您因"违规操作"
具体请查看互助需知
我知道了
×
提示
确定
请完成安全验证×
copy
已复制链接
快去分享给好友吧!
我知道了
右上角分享
点击右上角分享
0
联系我们:info@booksci.cn Book学术提供免费学术资源搜索服务,方便国内外学者检索中英文文献。致力于提供最便捷和优质的服务体验。 Copyright © 2023 布克学术 All rights reserved.
京ICP备2023020795号-1
ghs 京公网安备 11010802042870号
Book学术文献互助
Book学术文献互助群
群 号:604180095
Book学术官方微信