Construction And Comparison Of Fishing Ground Forecast Model Of Chub Mackerel(Scomber Japonicus) In Pacific Northwest

China has vigorously developed Pacific Northwest high seas fishery which is one of the pelagic fishery projects in recent years.In Pacific Northwest,Chub mackerel(Scomber japonicus)is one of the main fishing species.Lift net and falling net supplemented by lights are used for fishery operations.Because of the interaction and influence of the Kuroshio and western boundary stream from the Kuril,these interactions have formed the abundant resources of pelagic fish dominated by Chub mackerel.With the rapid development of fishery in the Northwest Pacific,the fishery forecasting technology in Pacific Northwest was also particularly important.Pelagic companies had more and more refined requirements for fishery forecasting technology.Pelagic fishing vessels required diversification and visualization of fishing ground information.The construction of fishery forecasting models based on machine learning could better meet the requirements of companies and fishing vessels.In this study,we first selected the historical catch data of Chub mackerel in the Pacific Northwest from 2014 to 2020 and obtained the Sea Surface Temperature(SST)data during the same period,which was obtained from NASA’s Moderate-Resolution Imaging Spectroradiometer(MODIS)tertiary product.Firstly,the temporal and spatial variation of Chub mackerel’s CPUE was analyzed with a spatial resolution of 0.2°×0.2°.At the same time,the change law of the center of gravity of Chub mackerel fishing grounds and the migration characteristics of Chub mackerel were studied by analyzing the inter-month variation of the center of gravity of Chub mackerel fishing grounds.The data set was constructed by Chub mackerel catch information and SST data.Thve data included data set preprocessing and data set enhancement.Subsequently,the GAM model,Alex Net and VGG16 convolutional neural network models were established for feature extraction.The application effects of the three models were verified and analyzed by using real fishery information.The comparison indicators included the number identified as correct,the number identified as wrong,the final model accuracy,and so on.On this basis,the influence experiment of the optimization parameters on the prediction effect of the convolutional neural network was completed.In general,the construction of the prediction model of the Chub mackerel fishery in the Pacific Northwest was completed.The results of the study are as follows:1)The spatial and temporal changes of Chub mackerel fishing grounds in the Pacific Northwest were studied,and the results were as follows: From 2014 to 2020,the annual average CPUE of Chub mackerel in the Pacific Northwest showed a decreasing trend as a whole.From March to August,the CPUE level was relatively low,and from September to November,the CPUE level was the highest in the whole year,which was the rich production period of Japanese mackerel in the Northwest Pacific.The average CPUE value was 17 tons/net.Both the longitude direction and the latitude direction had the phenomenon of turning back.The longitude direction showed a trend of changing from west to east and then back to the west.From September to November,the center of gravity of the fishery is more inclined to the northeast waters.2)A GAM model based on temporal and spatial factors and marine environmental factors SST was constructed,which used the historical production data of Chub mackerel in the Pacific Northwest from 2014 to 2019 and SST remote sensing data.The results showed that the training effect which was from training dataset divided into two groups was better than that without grouping according to the changing law of the center of gravity of the Chub mackerel fishery.The GAM model achieved an average Precision rate of 60.32%,Recall rate of 64.64%,and F1 score of 62.19%.The effect of predicting high-yield months(9-11)was better than that of low-yield months(4-8).It indicated that the GAM model had some validity in forecasting Chub mackerel fisheries in the Pacific Northwest.3)A dataset of four-channel information was constructed using Chub mackerel catch information and SST data,and the dataset was enhanced by random offset and rotation,which expanded the data by 6 times.This research constructed Alex Net model and VGG16 model based on deep learning principles for prediction of Chub mackerel fishing grounds.The Alex Net model and the VGG16 model finally achieved the highest test accuracy of 76.91% and 82.66%,respectively.The actual application effect of the model was verified by using the real Chub mackerel fishery information in 2020.The Alex Net model finally achieved an average Precision rate of 55.69%,a Recall rate of72.95%,and an average F1 score of 62.98%.The VGG16 convolutional neural network model finally achieved an average Precision rate of 58.45%,a Recall of 74.59%,and an average F1 score of 65.53%.On the whole,compared with the GAM model and the Alex Net model,the VGG16 model was more suitable for the construction of the Chub mackerel fishery forecast model in the Pacific Northwest.