Research On Dinoflagellate Red Tide Early Warning Model In Pingtan Coastal Aquiculture Area

Red tide has brought serious harm to marine ecology,economy and society.The causes of red tide outbreak are complex and many influencing factors make it difficult to predict it accurately.Therefore,taking the dinoflagellate red tide in Pingtan sea area of Fujian Province as the research object,this paper carries out the research on the construction and application of early warning model.By analyzing the historical monitoring data of red tide in Pingtan sea area,as well as the meteorological and water quality parameters and the temporal and spatial distribution law of dinoflagellate red tide in recent 10 years,this paper constructs an apparent oxygen increase early warning model based on mechanism analysis,and compares and studies five neural network algorithms,The non mechanism model of data-driven back propagation（BP）artificial neural network is constructed,and the model is verified by the monitoring data in April 2021.The following research results are obtained:In 2021,the main succession law of dominant phytoplankton in the breeding areas of Suao wharf and Liushui Wharf in Pingtan from late winter to spring to summer is diatom dinoflagellate diatom.The dominant species of dinoflagellates is Prorocentrum donghaiensis,The Dominant Algae in diatom are Chaetoceros debilis,Skeletonema costatum and Melosira sulcate.Apparent oxygen increment（AOI）can be used to measure the contribution of oxygen produced by phytoplankton photosynthesis to dissolved oxygen in seawater.Combined with the relationship between apparent oxygen increment and phytoplankton cell density,an early warning model of apparent oxygen increment is established.Models are established to fit K.mikimotoi,P.dentatum and various dominant dinoflagellates respectively,and the decisive coefficients and errors of each fitting formula are compared and analyzed,Finally,the fitting formula of apparent oxygen increase suitable for dinoflagellate red tide early warning in Pingtan sea area is determined as follows:=0.71237)2)-3.3468（R²=0.5857,n=60）.The fitting formula is verified by using the monitoring data in April 2021:that is,the prediction accuracy of apparent oxygen increase is 80%.According to the fitting formula,the threshold value of apparent oxygen increase applied to the early warning of dinoflagellate red tide in Pingtan sea area is 0.7 mg·L^-1,by launching the"citizen scientists for red tide observation"project,the algae density data collected by citizen scientists such as farmers are combined to improve the early warning accuracy of dinoflagellate red tide in Pingtan sea area.The marine monitoring data from 2013 to 2019 are input into five common artificial neural network models:k-nearest neighbor algorithm（KNN）regression model,random forest regression model,gradient lifting regression tree（GBRT）model,bagging regression model and BP artificial neural network model,and the accuracy is evaluated by comparing the output results with the measured values,The BP artificial neural network model with high accuracy is selected to construct the dinoflagellate red tide early warning model in Pingtan sea area.Using the BP artificial neural network red tide early warning calculus model,different combinations of meteorological and water quality factors are input into the model,and Chl-a and phytoplankton cell density are selected as the output ends of the model for calculus.When chlorophyll a is taken as the output index,the best model input combination is precipitation,air temperature,apparent oxygen increase and salinity.The decisive coefficient R² of the input combination is 0.782 and Mean absolute error（MAE）is0.163μg·L^-1,Root mean square error（RMSE）is 0.426μg·L^-1,the early warning accuracy reaches 83.7%.When phytoplankton cell density is used as the output of the model,the optimal input combination of the model is chlorophyll a,wind speed,p H and precipitation.The determination coefficient R² of the combination is 0.797,MAE is 0.081 cells·L^-1,RMSE is 0.142 cells·L^-1,and the early warning accuracy is 91.9%.There is a good correlation between the input index combination of the two groups of models,principal component analysis and Pearson correlation coefficient analysis,which confirms the reliability of the analysis results.Through the fuzzy probability analysis method,the indexes of chlorophyll a and phytoplankton cell density before and during the outbreak of dinoflagellate red tide in Pingtan sea area are analyzed.At the same time,combined with the previous relevant research results,the early warning values of chlorophyll a and phytoplankton cell density are set as 4.0μg·L^-1 and 3.0×10⁵cells·L^-1.The above research conclusions provide a reference basis for red tide early warning and prevention and control in Pingtan sea area.