基于海表温因子的太平洋褶柔鱼冬生群资源丰度预测模型比较

太平洋褶柔鱼是世界上重要的大洋性经济柔鱼类资源,其资源易受海洋环境因子的影响,科学预测其资源丰度有利于科学生产和管理。本实验依据2000—2010年太平洋褶柔鱼冬生群单位捕捞努力量渔获量(CPUE),以及产卵期间(1—3月)产卵场(28°~40°N、125°~140°E)的海表温(SST)数据,进行SST与CPUE的相关性分析,选取统计学有意义的SST作为影响资源丰度的因子,分别建立多元线性和BP神经网络的资源丰度预报模型,并利用2011和2012年的CPUE进行验证。结果显示,CPUE与产卵场1—3月SST相关系数较高的海域分别为1月的S1(30.5°N,136.5°E)和S2(31.5°N,136.5°E),2月的S3(30.5°N,137.5°E)和S4(30.5°N,135.5°E),3月的S5(37.5°N,129.5°E)和S6(37.5°N,130.5°E)。在多元线性及不同结构的BP神经网络等5种预报模型中,结构为6-4-1的BP神经网络模型预测精度最高,2011—2012年CPUE预测值精度平均为98%。研究表明,30°~32°N、135°~138°E和37°~38°N、129°~131°E附近海域的6个环境因子代表着1—3月产卵场暖流(黑潮和对马海流)势力的强弱,决定着当年太平洋褶柔鱼冬生群资源丰度,所建立的BP神经网络模型可作为其资源丰度的预测模型。

A comparative study on forecasting model of the stock abundance index for the winter-spawning cohort of Todarodes pacificus in the Pacific Ocean based on the factor of SST

Todarodes pacificus is one of important resources of the ocean economic Ommastrephidae in the world. In order to forecast the stock abundance of winter-spawning cohort, the catch per unit effort (CPUE) as abundance index from T. pacificus stock assessment report of Japan in 2013 is used to establish the forecasting model in this study. The correlation analysis between sea surface temperature (SST) in the spawning areas of 28°N-40°N and 125°E-140°E and CPUE from January to March during 2000-2010 was carried out respectively to select the significantly affecting factors in statistics. The multivariate linear model and BP neural network model forecasting abundance index of T. pacificus winter-spawning population were established and compared, and the actual CPUE in 2011 and 2012 was used for validation. The results showed that the spawning areas with high correlation coefficient between CPUE and SST in Jan. to Mar. are S1 (30.5° N, 136.5° E) and S2 (31.5° N, 136.5° E) in January, the correlation coefficient are 0.71 and 0.70 respectively; S3 (30.5° N, 137.5° E) and S4 (30.5° N, 135.5° E)in February, and the correlation coefficient are 0.87 and 0.84, respectively; S5 (37.5° N, 129.5° E) and S6 (37.5° N, 130.5° E) in March, and the correlation coefficient are 0.72 and 0.70, respectively. Total of five forecasting models including multivariate linear model and BP neural network model with different structure are established and compared. The BP 6-4-1 neural network model is the best, and the average prediction accuracy of the CPUE value during 2011-2012 attained 98%. This study suggests that the model can be used as the forecasting model of the stock abundance for T. pacificus winter-spawning cohort.