基于GAM和BRT模型的不同渔汛期水温垂直结构对西北太平洋柔鱼CPUE的影响

Impact of vertical structure of water temperature during different fishing seasons on CPUE of neon flying squid(Ommastrephes bartramii)in the Northwest Pacific Ocean using GAM and BRT models

为探究水温对西北太平洋柔鱼(Ommastrephes bartramii)单位捕捞努力量渔获量(CPUE)的影响,利用2015—2019年间5—11月西北太平洋柔鱼的生产统计数据,结合同时期海洋环境数据,对不同渔汛阶段分别采用广义加性模型(generalized addictive models,GAM)及提升回归树(boosting regression tree,BRT)模型,分析不同水层(0-300 m)温度及垂直温度梯度对CPUE的影响,并采用多次十折交叉验证评估了两种模型的稳定性和准确性。结果表明:渔汛初期(5—7月),GAM模型筛选的关键因子依据偏差解释率大小依次为纬度(Lat)和150 m水层温度(T_(150))、50-100 m水温梯度(G_(50-100))、经度(Lon)和100-150 m水温梯度(G_(100-150)),BRT模型筛选出的关键因子按贡献率大小依次为表层温度(T_(0))、0-50 m水温梯度(G_(0-50))、50 m水层温度(T50)、Lat和G_(50-100);渔汛旺期(8—11月),GAM模型筛选出的关键因子依据偏差解释率大小依次为Lat、Lon、T_(0)、G_(0-50)和G_(50-100),BRT模型筛选出的关键因子按贡献大小率依次为G_(0-50)、Lon、T_(0)、Lat和T50;两种模型比较显示,BRT模型在拟合优度方面优于GAM模型。研究表明,不同渔汛阶段西北太平洋柔鱼CPUE的关键影响因素有所不同,渔汛初期中上层水温对CPUE影响较大,而渔汛旺期浅水层水温对CPUE贡献较大。

To explore the effect of water temperature at different depths and vertical structure of water temperature on catch per unit effort(CPUE)of neon flying squid(Ommastrephes bartramii)and to provide a guidance for the improvement of neon flying squid production efficiency,the influence of 0-300 m water temperature and vertical structure at different depths on the CPUE were analyzed based on the fishery data from May to November from 2015 to 2019,combined with the marine environmental data of the same period,the generalized additive model(GAM)and the boosting regression tree(BRT)during different fishing seasons.The stability and accuracy of the two models were evaluated using repeated ten-fold cross-validation.The results showed that the key factors screened by the GAM model based on the magnitude of deviation explained were found to be latitude(Lat),150 m layer temperature(T_(150)),50-100 m water temperature gradient(G_(50-100)),longitude(Lon),and 100-150 m water temperature gradient(G_(100-150))during the early fishing season(May to July).The BRT model screened factors in order of contribution rate were shown to be sea surface temperature(T_(0)),0-50 m water temperature gradient(G_(0-50)),50 m layer temperature(T50),Lat,and G_(50-100).During the main fishing season(August to November),however,the key factors selected by the GAM model based on the magnitude of deviation explained were Lat,Lon,T_(0),G_(0-50),and G_(50-100).The BRT model selected factors in order of contribution rate were G_(0-50),Lon,T_(0),Lat,and T50.The comparison of the two models indicated that BRT was better than the GAM to analyse the actual fishing data,with different key factors affecting CPUE in different stages of fishing season.The great influence on CPUE was observed in the upper-middle water temperature in the early fishing season,and great contribution to CPUE was found in the shallow water temperature in the main fishing season.