Application Of Machine Learning Method In Estimating The Biomass Of Main Economic Crabs In Zhoushan Fishery

Fishery resources assessment technology is an important tool to understand the information of Marine fishery resources,formulate fishery management plans and ensure the sustainable utilization of Marine fishery resources.It can not only protect the ecological environment,but also provide a basis for the formulation of fishery policies and help achieve the economic,social and environmental objectives of fishery management.For a long time,the area sweeping method has been widely used in fishery biomass assessment.Its main advantages are simple calculation and strong operability.However,this method needs to assume uniform distribution of resources,and to improve the accuracy of biomass assessment,it is necessary to improve the survey scheme and increase the survey budget.Nowadays,fishery resources assessment technology is no longer limited to the traditional methods of obtaining fish data,but flexibly uses diversified machine learning models and statistical algorithms,and applies new technologies such as remote sensing technology and big data technology to obtain a large number of data from satellite images and other sensors,which improves the scientific nature and accuracy of fishery resources assessment.It provides scientific,accurate and comprehensive basis for fishery management and protection,In this study,based on the data of various economic crab species obtained from the bottom trawl survey of fishery resources in Zhoushan fishing grounds in August 2006 and January,May and November 2007,we studied the fishery resources of Portunus trituberculatus,Charybdis bimaculata,Japanese sturgeon,and Charybdis japonica in Zhoushan fishing grounds.The biomass of four major economic crab species,Portunus trituberculatus,Charybdis bimaculata,Charybdis japonica and Ovalipes punctatus,was compared with the random forest(RF),gradient boosting regression tree(GBRT)and extreme gradient boosting regression tree(EGRT).GBRT and Extreme gradient boosting(XGBoost).The relationship between the spatial and temporal distribution of four economic crab species and environmental factors in Zhoushan fishery was analyzed,while the best-fit model was screened by variance explained rate(VE)and other indicators.The biomass distribution of the remaining stations was assessed by the established models,and the simulations were repeated 20,000 times.Finally,all the biomass of the four economic crab species at 20 stations in the surveyed sea area was estimated by RF,GBRT and XGBoost 3 methods,and the results estimated by the swept area method were compared to compare the difference in accuracy between different assessment methods,and the assessment results were found to be closer to the real values The method was used to compare the accuracy of different assessment methods and to find the assessment results that are closer to the real value.The following results were obtained from the study.1.In this study,the biomass of four economic crab species was estimated by reducing the number of input modeling stations and establishing the optimal models according to the nature of different models,comparing the accuracy and practicality of the four crab fishery resource assessment methods,and finding that factors such as bottom temperature and pH had significant effects on the distribution of several economic crab species,and concluding that with the reduction of the number of survey stations,the XGBoost method was more effective in estimating the biomass of four economic crab species under the conditions of unfocused and fluctuating data,such as autumn and winter.The XGBoost method was found to be significantly better than the swept area method in assessing biomass as the number of survey stations decreased,and the error was reduced by 7.49%～21.34%under unfocused and fluctuating data conditions,such as autumn and winter seasons.2.By estimating the biomass of four economic crab species through modeling with uniform data distribution,it was concluded that the difference between the results of the swept sea area method and the machine learning method was not significant(P>0.05)in the case of concentrated and less fluctuating data,such as spring and summer.3.After comparing the estimation results of three machine learning algorithms,it was concluded that all three machine learning algorithms were able to predict the biomass of major economic crab species in Zhoushan fishery better,among which the XGBoost algorithm had the best prediction effect,which was consistent with the results of other studies.4.In this study,the machine learning method was used to estimate the biomass of main economic crabs in Zhoushan fishery.The results showed that the machine learning method had better estimation accuracy and generalization ability than the traditional method.Especially in the consideration of various factors and feature screening,machine learning method has obvious advantages.It is more scientific and efficient to estimate the biomass of main economic crab resources in Zhoushan fishery by using machine learning method,which provides a new way to estimate the biomass of crab resources in the future.This study analyzed the relationship between spatial and temporal distribution of four economic crab species and environmental factors based on multiple models,explored the application of machine learning methods in biomass assessment,with the aim of improving assessment accuracy and saving resource survey costs,and provided new ideas for future exploration of biomass estimation of major economic crab species in Zhoushan sea area through the process and methods of this study.