基于LSTM模型的中西太平洋鲣栖息地预测  

Habitat prediction of skipjack in the Western and Central Pacific based on LSTM model

作　　者：周成[1,2,3,4] 周想 胡媛媛[1] 刘力文 ZHOU Cheng;ZHOU Xiang;HU Yuanyuan;LIU Liwen(College of Marine Living Resource Sciences and Management,Shanghai Ocean University,Shanghai 201306,China;National Engineering Research Center for Oceanic Fisheries,Shanghai 201306,China;Key Laboratory of Sustainable Exploitation of Oceanic Fisheries Resources,Ministry of Education,Shanghai 201306,China;Key Laboratory of Ocean Fisheries Exploration,Ministry of Agriculture and Rural Affairs,Shanghai 201306,China)

机构地区：[1]上海海洋大学海洋生物资源与管理学院,上海201306 [2]国家远洋渔业工程技术研究中心,上海201306 [3]大洋渔业资源可持续开发教育部重点实验室,上海201306 [4]农业农村部大洋渔业开发重点实验室,上海201306

出　　处：《上海海洋大学学报》2025年第1期153-163,共11页Journal of Shanghai Ocean University

基　　金：国家重点研发计划(2023YFD2401301);农业农村部全球渔业资源调查监测评估(公海渔业资源综合科学调查)专项(D-8025-23-1003)。

摘　　要：为了解决传统栖息地预测模型中无法捕捉具有时间序列信息的环境因子对金枪鱼空间分布滞后影响的不足。采用2021—2024年金枪鱼围网渔捞日志数据,通过构建滞后天数为1、5、10、15 d的长短期记忆(Long-short term memory,LSTM)神经网络模型,分别对单位捕捞努力量渔获量(Catch per unit of effort,CPUE)和经纬度进行了预测。研究表明,滞后10 d的模型精度最高,其均方误差(Mean square error,MSE)为0.018 7,平均绝对误差(Mean absolute error,MAE)为0.077 6,表明鲣空间分布受过去短期内环境累计效应的影响。通过对最佳模型进行验证,结果表明预测纬度与实际纬度之间的R2为0.97,预测经度与实际经度之间的R2为0.65,说明空间分布预测范围与实际基本吻合。为揭示鲣栖息地特征及其生态过程的动态机制提供了新的理解,同时为中西太平洋鲣围网渔业的科学管理提供了重要参考依据。To address the limitations of traditional habitat prediction models in capturing the lagged effects of environmental factors with time series information on tuna spatial distribution,this study utilized tuna purse-seine fishing log data from 2021 to 2024.Long-short term memory(LSTM)neural network models were constructed with lag durations of 1 day,5 days,10 days,and 15 days to predict catch per unit of effort(CPUE)and geographic coordinates(latitude and longitude).The findings indicate that the 10-day lag model exhibited the highest accuracy,with a mean square error(MSE)of 0.0187 and a mean absolute eror(MAE)of 0.0776,suggesting that the spatial distribution of skipjack is influenced by cumulative short-term environmental effects.Validation of the optimal model demonstrated the R’of 0.97 for predicted versus actual latitude and 0.65 for longitude,indicating a strong alignment between predicted and observed spatial distributions.This research offers new insights into the dynamic mechanisms underlying skipjack tuna habitat characteristics and ecological processes.Furthermore,it provides critical references for the scientific management of skipjack purse seine fisheries in the Western and Central Pacific Ocean.

关 键 词：鲣 栖息地预测 长短期记忆模型 中西太平洋 

分 类 号：S931.1[农业科学—渔业资源]