Analysis Of Temporal And Spatial Distribution Of Portunus Trituberculatus In The Northern East China Sea Based On Multiple Models

Portunus trituberculatus is an economically important crab and one of the most important fishery species caught in the East China Sea.In recent years,with the increase in fishing pressure and changes in marine environment,the yield of P.trituberculatus has considerably fluctuated,and its resources are showing signs of overfishing.Relevant government departments have taken measures to ensure the sustainability of P.trituberculatus resources,such as by increasing the intensity of stock enhancement and by extending the fishing moratorium period for P.trituberculatus.In 2017,the Ministry of Agriculture and Rural Affairs officially launched a pilot project on P.trituberculatus quota fishing in the northern Zhejiang fishery.An accurate understanding of the resource size and distribution of P.trituberculatus is a prerequisite for this work,and only by fully understanding the current status of these resources can a more reasonable management strategy be developed.In this study,data on the fishery resources and environmental distribution of P.trituberculatus were collected,spanning four seasons from 2006 to 2007.The relationship between the spatial and temporal distribution of P.trituberculatus and environmental factors in the northern East China Sea was analyzed using six models,namely,generalized additive model(GAM),support vector machine(SVM),artificial neural network(ANN),random forest(RF),gradient boosting regression tree(GBRT),and extreme gradient boosting(XGboost).Best-fit models were screened using Akaike information criterion and percentage of variance explained by the model.Overfitting of the different models was discerned using the~2 difference between model fitting and model prediction(Δ~2)as an indicator.The following conclusions were drawn:1.Five of the six models in spring provided the greatest contribution of sea-bottom temperature(SBT)to the model.Similarly,three models in summer and autumn gave the greatest contribution of SBT to the model.Finally,four of the six models in winter presented the greatest contribution of p H to the model.Moreover,F-test results indicated that SBT and p H had significant effects on the distribution of relative resource density of P.trituberculatus(P<0.05).2.According to the cumulative explanatory rate of the models as an indicator to compare the fit of the models between different seasons,the fit of the models in spring was better than that of the models in other seasons.The fit of the XGBoost models was the best in all four seasons,followed by that of the ANN models.By contrast,the fit of the GBRT models was relatively poor.3.Comparison of the overfitting of the different models using(?)~2 as an indicator revealed that the overfitting of the spring models was the least severe,with(?)~2 values below 0.3,whereas the overfitting of the autumn models was the most severe,with (?)~2 values greater than 0.51.The three single-structure models,namely,GAM,ANN,and SVM,were more severely overfitted than the three integrated learning models,namely,RF,GBRT,and XGBoost.4.Comparison of the predictive performance of the different models in terms of coefficient of determination (R~2) and relative root-mean-square-error revealed that the spring models outperformed the models of the three other seasons in terms of predictive performance.In spring and winter,the RF models had the best predictive performance,whereas in summer and autumn,the XGBoost models had the best predictive performance.The SVM models had the worst predictive performance in all seasons,except autumn.5.The optimal predictive model for each season was used to predict the distribution of P.trituberculatus resources in the northern East China Sea.Results showed notable differences in the distribution of P.trituberculatus resources in different seasons,with a more even distribution and a higher average resource density in summer than those in the other seasons,a higher resource density in the northeast than in the southwest in spring,and a higher resource density in the east than in the west in autumn.In winter,the resource density of P.trituberculatus was relatively high in the central and northern regions,but it was low in the southeastern region.This study analyzed the relationship between the spatial and temporal distribution of P.trituberculatus and environmental factors based on various models.Results showed that SBT and p H had remarkable effects on the distribution of P.trituberculatus.A seasonal predictive model for the distribution of P.trituberculatus resources was proposed in consideration of the advantages and disadvantages of the different models.This predictive model will allow a more accurate prediction of the distribution of P.trituberculatus resources in the northern East China Sea,and it offers guidance for the rational exploitation and scientific management of these resources.