Scholarlink

<p>An effective fishery forecast model contributes to the efficient conservation and management of albacore tuna (<i>Thunnus alalunga</i>). Given the emerging use of long short-term memory (LSTM) models and quantile regression (QR) models for prediction in fishery, this study evaluated their performances in predicting the catch per unit effort (CPUE) for albacore tuna. On the basis of the electronic logbook systems (ELBS) data of a Chinese distant-water fishing company, spanning from 1 January 2017 to 31 May 2021, nominal CPUE was calculated. The spatial resolution was set as 1° × 1°, and the temporal resolution was daily. After screening the spatiotemporal and environmental factors as explanatory variables, LSTM and QR models were developed and compared the performance for CPUE prediction. The results show the following: (1) The LSTM model demonstrates superior prediction ability compared with the QR model. (2) The LSTM model provides more accurate predictions in high CPUE areas. (3) It is necessary to conduct correlation analysis and multicollinearity diagnosis on variables before modeling. (4) Key factors significantly impacting the CPUE prediction of both LSTM and QR models include sea surface temperature, chlorophyll-<i>a</i> concentration, and temperature at the 200&#xa0;m water layer. For the LSTM model, dissolved oxygen concentrations at 150&#xa0;m and 200&#xa0;m are also identified as important variables, while the QR model emphasizes the contribution of interaction terms between variables to CPUE prediction. These findings suggest that the LSTM model is a reliable tool for predicting tuna CPUE.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Comparison of LSTM and QR models for predicting CPUE of albacore tuna in waters near the Cook Islands

  • 期刊论文

  • 发表日期：

    2025年2月7日

查看全文
  • Hui Xu,
  • Liming Song,
  • Yuwei Li,
  • Tianjiao Zhang,
  • Jieran Shen

摘要

An effective fishery forecast model contributes to the efficient conservation and management of albacore tuna (Thunnus alalunga). Given the emerging use of long short-term memory (LSTM) models and quantile regression (QR) models for prediction in fishery, this study evaluated their performances in predicting the catch per unit effort (CPUE) for albacore tuna. On the basis of the electronic logbook systems (ELBS) data of a Chinese distant-water fishing company, spanning from 1 January 2017 to 31 May 2021, nominal CPUE was calculated. The spatial resolution was set as 1° × 1°, and the temporal resolution was daily. After screening the spatiotemporal and environmental factors as explanatory variables, LSTM and QR models were developed and compared the performance for CPUE prediction. The results show the following: (1) The LSTM model demonstrates superior prediction ability compared with the QR model. (2) The LSTM model provides more accurate predictions in high CPUE areas. (3) It is necessary to conduct correlation analysis and multicollinearity diagnosis on variables before modeling. (4) Key factors significantly impacting the CPUE prediction of both LSTM and QR models include sea surface temperature, chlorophyll-a concentration, and temperature at the 200 m water layer. For the LSTM model, dissolved oxygen concentrations at 150 m and 200 m are also identified as important variables, while the QR model emphasizes the contribution of interaction terms between variables to CPUE prediction. These findings suggest that the LSTM model is a reliable tool for predicting tuna CPUE.