Spatial And Temporal Distribution Of Skipjack Tuna (Katsuwonus Pelamis) Resources In The Western And Central Pacific Ocean Based On Multi-Output Neural Networks

Marine fishes have high economic and nutritional values,and the utilization and conservation of marine fishery resources have become an important direction of marine fishery research.The changes in marine environment caused by large-scale climatic phenomena and human activities will have a significant impact on the spatio-temporal distribution of marine fish resources,and it is necessary to conduct research on the changes in the distribution of marine fish and their response to the changes in marine environment.With the development and wide application of technologies such as machine learning and big data,fish distribution research has entered a new stage.Artificial neural network models,as emerging models,outperform traditional model methods in screening and abundance prediction of important environmental factors affecting fish distribution.Biological data are high heterogenous with multiple levels and complex structures have been studied extensively in ecology and evolution.Increasing studies have shown that heterogeneity of time,space,fishing operation,and environments also exist in fishery dependent data.Skipjack tuna has a crucial role in the global tuna fishery,and its fishing method is mainly purse seine operation.According to the cluster properties of the species,it can be generally divided into two categories: free school and ancillary fish school.With the continuous improvement of fishing technology,Fish Aggregation Devices(FAD)have started to be used increasingly in the fishing methods of marine fishes.Skipjack tuna has a wide variety and complexity of fishing methods,and its fishery data has a high heterogeneity.Therefore,the standardization and prediction of heterogeneous skipjack tuna fishery data requires more advanced technical methods and fuller consideration of heterogeneity factors.This study explored the spatial and temporal variation patterns of the center of gravity of skipjack tuna fisheries based on production data of skipjack tuna purse-seine fishing vessels and marine environmental factors in the central and western Pacific Ocean from 1995 to 2018.The heterogeneous skipjack tuna fishery data were also standardized to predict the location and resource abundance of skipjack tuna fishery in the medium and long term.Marine environmental data were used as input factors,including Sea Surface Temperature(SST),Sea Surface Salinity(SSS),Dissolved Oxygen(O2),Primary Productivity(PP)and ENSO(Nino3.4),as well as the production data of four types of skipjack tuna as multiple output factors,to construct a multi-output neural network model.The optimal parameters of the model were searched by manual grid search method,and the multi-output neural network model with the implicit layer node of 13 was finally the optimal model.The specific results are as follows.(1)The four fishing types of skipjack tuna in the western and central Pacific Ocean are mainly distributed spatially in the tropical waters near the equator.The center of gravity of the free schools is distributed from 140°E to 160°W in longitude and from5°N to 10°S in latitude;the center of gravity of the log schools is distributed from140°E to 170°W in longitude and from 5°N to 8°S in latitude;the center of gravity of the drifting fishing artificial devices is relatively concentrated and distributed from140°E to 160°W in longitude and from 3°N to 8°S in latitude;the center of gravity of the drifting fishing artificial devices is relatively concentrated and distributed from140°E to 160°W in longitude and from 3°N to 8°S in latitude;the center of gravity of the fishery of the artificial fishing aggregating devices was the most concentrated than the former schools,with little monthly variation,and was concentrated in the waters off the coast of Papua New Guinea.The results of K-means clustering showed that the center of gravity of the fishery shifted to the west when the SSTA value was low,while the center of gravity of the fishery shifted to the east when the SSTA value was high except for the artificial fishing aggregating devices.The spatial and temporal variation of the center of gravity of skipjack tuna fishery was significantly influenced by the ENSO phenomenon.(2)According to the results of the multi-output neural network model,the values of both evaluation indexes RMSE and MAE fluctuate with the number of nodes in the hidden layer.We chose the model with the number of nodes in the hidden layer of 13 as the optimal model for evaluating the bonito resources in the western and central Pacific Ocean from 1995-2018.SST is the most important variable affecting the abundance of skipjack tuna resources.The marine environmental factors were SST,PP,Nino3.4,O2 and SSS in descending order of importance,while the other factors were Lat,Month and Lon in descending order of importance.(3)The areas with higher abundance of skipjack tuna in the western and central Pacific Ocean predicted by using the multi-output neural network model were mainly concentrated between 10°S~10°N and 140°E~180°,while the spatial distribution of different abundances had some differences.The changes in nominal CPUE and standardized CPUE were not obviously consistent in time series,but the trends were somewhat correlated and lagged,mainly due to the influence of the ENSO phenomenon.This study provides a new theoretical and modeling support for the study of the spatiotemporal distribution and fishery forecasting of skipjack tuna resources,and provides new ideas for the spatio-temporal distribution of marine fishes.