A Comparison Of Yellowfin Tuna Fishing Condition Forecast Models Based On Bayesian Method,Quantile Regression And Artificial Neural Network Model

Yellowfin tuna(Thunnus albacares)is one of the important catch species in world tuna fisheries,and it’s also one of the important fisheries resources in the distant-water tuna longline fisheries of China.So it is necessary to study in depth to achieve efficient and sustainable use of the resources.In order to quantify its’ abundance and resource status better,we need to use the environment data and catch data in high resolution,we can predict effectively the integrated habitat index(IHI)model of yellowfin tuna.Because of the close relationship between the distribution of the resources and environmental factors,it is necessary to apply the related environmental factors to the study of predicting CPUE and integrated habitat index for yellowfin tuna.In this study,we used the environment data(including spatial resolution of 0.25 °×0.25°sea surface temperature,sea surface height,primary productivity)and the corresponding yellowfin tuna catch data(including the operation time,position,the number of being capture and the number of hooks of each day,each ship)to build predicting model of catch per unit fishing effort(CPUE)by Bayesian Analysis(BA),Quantile Regression Model(QRM)and Artificial Neural Networks(ANNS).By these models,we can get the predicting CPUE,and calculate the integrated habitat index of three models and evaluating whether it is significant correlation with the measured CPUE.We can compare the ability of three predicting models.In addition,we input the environment data of 110 verification sites,into the three models one by one,we got the corresponding predicting CPUE,and we evaluated whether it is significant correlation with the measured CPUE by Wilcoxon rank test,and identified the best model.The results showed that:1)The IHI value of three models was different.The predication ability of QRM model is the best especially in the high IHI areas,and the prediction model based on ANNS is the second.The prediction model based on BA is the third.2)The areas with the high IHI of yellowfin tuna obtained from BA model,QRMmodel and ANNS model were different from each other.The areas with the high IHIof yellowfin tuna obtained from BA model,QRM model and ANNS model was in thearea of 0o00’N ~ 4o00’N,150o00’E ~ 165o00’E;2o 00’N ~ 9o 00’N,153o00’E ~ 165o00’E;and 1o00 ’ N ~ 7o00’N,152o30’E ~ 165o30’E;respectively.The high IHI area foryellowing tuna of three models covered the area of 2o00’N ~ 4o00’N,153o00’E ~165o00’E.3)The occurred probability of fishing ground,which predicted by BA,and the predicting CPUEs,which predicted by QRM and ANNS models were obtained by inputting the environment data(accounted for 14.3% of the total data)of 110 verifying sites.The correlation between the predicting data and the measured data was analyzed by Wilcoxon rank test and the predicting ability of three models were evaluated.The predicting model based on QRM model was the best,the model based on ANNS was the second,and the model based on BA was relatively weak.The correlation between the predicting data and the measured data was analyzed by EF value and the predicting ability of three models were evaluated.The predicting model based on BA model was the best,the model based on QRM was the second,and the model based on ANNS was relatively weak.4)The key environmental variables,which influence the yellowfin tuna distribution were different from the models.The key environmental variable of BA was sea surface temperature,the key environmental variables of QRM were sea surface temperature,sea surface height,primary productivity and their interactions,while the key environmental variables of ANNS were sea surface height and month.5)The model application were different from each other.The predicting model based on QRM was good at choosing the key environmental variables.The disadvantage was this model will reject more environmental variables if they were weakly related to the CPUE.The predicting model based on ANNS was suit for building the model using multi environmental variables but it can not be indicated in equation except the predicting value.The predicting model based on BA was good at predicting the occurring probability of the central fishing grounds.The predicting ability was low in the high value area and the predicting result was relatively even.