| Tunas are widely distributed in tropical and subtropical waters of the Pacific,Atlantic and Indian oceans.Among them,Thunnus alalunga,Thunnus obesus and Thunnus albacares are the main target fish species of tuna longline fishery in China.The three tunas are highly migratory species,and the formation of their fishing grounds is closely related to environmental factors.Therefore,accurate prediction of the location of the central fishing ground can greatly reduce the time to find fishing grounds and improve productivity,which plays an important role in tuna fisheries.In order to understand the effects of different models and spatial resolutions on the prediction accuracy of tuna fishing grounds in tropical Atlantic waters,the fishing logbook data of 13 Chinese longline fishing vessels operating in the high seas of the Atlantic from 2016 to 2019 were collected.Twenty nine marine environmental factors(sea surface wind speed,chlorophyll-a concentration,eddy kinetic energy,depth of mixed layer and vertical temperature,salinity and dissolved oxygen of 500 m water layer)were selected.The correlation between CPUE of Bigeye tuna,Yellowfin tuna and Albacore tuna and 29 marine environmental factors was analyzed,and the environmental factors related to CPUE were screened out.Then the collinear analysis of these environmental factors was carried out,and the non-collinear marine environmental data were further selected for modeling.Based on five models(random forest(RF),gradient lifting decision tree(GBDT),k nearest neighbor(KNN),logistic regression(LR)and stacking integration(integrated by RF,GBDT and KNN models,STK),the optimal parameters were determined by Grid Search and 50% discount cross validation with day as temporal resolution and with spatial resolutions of 0.5 °× 0.5 °,1 °× 1 °,2 °× 2 ° and 5 °× 5 °.The data from 25% of the sites were substituted into the model as test data for verification.The results of 5 models and 4 resolutions were evaluated according to the prediction accuracy the comprehensive evaluation index F1 score and F1 variation rate.The results showed that:(1)For Bigeye,Yellowfin and Albacore tuna longline fisheries in the tropical Atlantic Ocean,the accuracy and F1 score of STK model were high;In the STK model,the spatial resolution with the highest prediction accuracy and F1 score of Bigeye,Yellowfin and Albacore tuna was 1°×1°,5°×5° and 5° ×5°,respectively;For Bigeye tuna,when the spatial resolution was 0.5°×0.5°,the accuracy of RF was 79.74%,and the F1 score was 85.61%;the accuracy of GBDT was 79.95%,and the F1 score was 85.68%;the accuracy of KNN was 79.70%,and the F1 score was85.73%;the accuracy of LR was 75.26%,and the F1 score was 82.93%;the accuracy of STK model was 80.27%,and the F1 score was 85.92%.When the spatial resolution was 1°×1°,the accuracy of RF was 86.08%,and the F1 score was 89.91%;the accuracy of GBDT was 85.60%,and the F1 score was 89.40%;the accuracy of KNN was83.70%,and the F1 score was 87.38%;the accuracy of LR was 74.25%,and the F1 score was 82.39%;the accuracy of STK model was 86.31% and the F1 score was89.84%.When the spatial resolution was 2°×2°,the accuracy of RF was 79.15%,and the F1 score was 84.80%;the accuracy of GBDT was 79.60%,and the F1 score was85.11%;the accuracy of KNN was 79.35%,and the F1 score was 84.35%;the accuracy of LR was 75.19%,and the F1 score was 82.33%;the accuracy of STK model was80.30%,and the F1 score was 85.52%.When the spatial resolution was 5°×5°,the accuracy of RF was 78.95%,and the F1 score was 84.56%;the accuracy of GBDT was79.03%,and the F1 score was 84.44%;the accuracy of KNN was 77.44%,and the F1 score was 83.35%;the accuracy of LR was 74.27%,and the F1 score was 81.73%;the accuracy of STK model was 79.20%,and the F1 score was 84.68%.For Yellowfin tuna,when the spatial resolution was 0.5°×0.5°,the accuracy of RF was 66.26%,and the F1 score was 66.43%;the accuracy of GBDT was 65.30%,and the F1 score was 65.85%;the accuracy of KNN was 63.56%,and the F1 score was60.68%;the accuracy of LR was 63.17%,and the F1 score was 65.42%;the accuracy of STK model was 66.58%,and the F1 score was 66.68%.When the spatial resolution was 1°×1°,the accuracy of RF was 65.78%,and the F1 score was 98.53%;the accuracy of GBDT was 64.83%,and the F1 score was 67.78%;the accuracy of KNN was64.75%,and the F1 score was 64.94%;the accuracy of LR was 60.92%,and the F1 score was 65.33%;the accuracy of STK model was 65.90%,and the F1 score was68.63%.When the spatial resolution was 2°×2°,the accuracy of RF was 68.72%,and the F1 score was 72.70%;the accuracy of GBDT was 68.22%,and the F1 score was72.31%;the accuracy of KNN was 67.02%,and the F1 score was 69.76%;the accuracy of LR was 63.61%,and the F1 score was 68.60%;the accuracy of STK model was69.12%,and the F1 score was 72.86%.When the spatial resolution was 5°×5°,the accuracy of RF was 72.45%,and the F1 score was 79.68%;the accuracy of GBDT was71.45%,and the F1 score was 78.94%;the accuracy of KNN was 70.62%,and the F1 score was 78.40%;the accuracy of LR was 69.95%,and the F1 score was 78.82%;the accuracy of STK model was 72.95%,and the F1 score was 79 78%.For Albacore tuna,when the spatial resolution was 0.5°×0.5°,the accuracy of RF was 72.88%,and the F1 score was 77.19%;the accuracy of GBDT was 72.09%,and the F1 score was 76.67%;the accuracy of KNN was 72.88%,and the F1 score was77.63%;the accuracy of LR was 69.14%,and the F1 score was 74.52%;the accuracy of STK model was 73.41%;the F1 score was 77.83%.When the spatial resolution was1°×1°,the accuracy of RF was 73.37%,and the F1 score was 78.73%;the accuracy of GBDT was 73.25%,and the F1 score was 78.56%;the accuracy of KNN was 72.10%,and the F1 score was 76.87%;the accuracy of LR was 70.48%,and the F1 score was76.84%;the accuracy of STK model was 73.61%;the F1 score was 78.83%.When the spatial resolution was 2°×2°,the accuracy of RF was 76.49%,and the F1 score was81.76%;the accuracy of GBDT was 74.04%,and the F1 score was 80.09%;the accuracy of KNN was 75.19%,and the F1 score was 80.36%;the accuracy of LR was70.63%,and the F1 score was 78.09%;the accuracy of STK model was 76.94%,and the F1 score was 82.14%.When the spatial resolution was 5°×5°,the accuracy of RF was 77.55%,and the F1 score was 84.60%;the accuracy of GBDT was 77.30%,and the F1 score was 85.07%;the accuracy of KNN was 75.29%,and the F1 score was82.79%;the accuracy of LR was 72.70%,and the F1 score was 82.63%;the accuracy of STK model was 78.21%,and the F1 score was 84.96%.For Bigeye tuna and Yellowfin tuna,the spatial resolution with the highest F1 variation rate was 1 °× 1 °;For Albacore tuna,the spatial resolution with the highest F1 variation rate was 2 °× 2 °.(2)The CPUE of Bigeye tuna was closely related to the dissolved oxygen concentration of 200 ～ 300 m water layer,salinity and temperature of 500 m water layer.(3)The CPUE of Yellowfin tuna was correlated with the dissolved oxygen concentration in 200 m and 500 m water layers,and the surface water temperature and the salinity of 500 m water layer would also affect the distribution of CPUE of Yellowfin tuna.(4)The CPUE of Albacore tuna was closely related to salinity,dissolved oxygen concentration and temperature of 100 m water layer,which would affect the distribution of CPUE of Albacore tuna.This study suggests:(1)Using STK model to predict the fishing grounds of Bigeye,Yellowfin and Albacore tuna in the tropical Atlantic Ocean;(2)It is necessary to analyze the correlation between the CPUE of Bigeye,Yellowfin and Albacore tuna and relevant marine environmental factors to delete the unrelated environmental factors;(3)It is necessary to diagnose the marine environmental factors by collinearity and delete the environmental factors with collinearity;(4)When making fishery forecasts for Bigeye,Yellowfin and Albacore tuna fishing grounds in the tropical Atlantic Ocean,the spatial resolutions were taken as1°×1°,1°×1° and 2°×2°,respectively,to obtain the higher fishing grounds forecast accuracy and optimum spatial resolution. |
PDF Full Text Request