Associated Factor Analysis Of The Occurrence Degree Of Sitobion Avenae On Winter Wheat In Eastern Shandong Province

Wheat(Triticum aestivum L.)is one of the main food crops in China,and Sitobion avenae F.is the dominant pest species of wheat plant.S.avenae are typical R pests,which have a strong ability to reproduce and migrate.In recent years,the overwintering rate of aphids has increased,and migration time is early due to global climate change.This further strengthened that wheat aphids are difficult to be controlled.The monitoring and forecasting of aphids,especially the short-term monitoring and forecasting,is impossible to realize.By using a large amount of data,the monitoring and forecasting model for S.avenae was constructed.The key factors affecting the occurrence level of the aphid were analyzed,and short-term monitoring and forecasting models were conducted.This can guide people to develop scientific and efficient control strategies of this pest.This research is based on the concept of big data.First,we collected and analyzed the occurrence data of S.avenae and 17 environmental variables such as lady beetles and average temperature in eastern Shandong Province from 2003 to 2019.Second,we build a S.avenae prediction model using random forest,support vector machine and neural network algorithm.At the same time,we use confusion matrix and ten-fold cross-validation methods to evaluate and select the optimal model.Finally,we used the best model to analyze the association between the occurrence of S.avenae and environmental factors.It can be concluded from the confusion matrix that the accuracy of the random forest model,support vector machine model,and neural network model classification are 0.84,0.83 and 0.77,respectively.It can be concluded from ten-fold cross validation that the areas under the receiver operating characteristic curve(AUC)of the random forest model,support vector machine model,and neural network model are 0.88,0.87 and 0.82,respectively.In terms of sensitivity,the average sensitivity of the random forest model,the support vector machine model and the neural network model are 0.92,0.99 and 0.88,respectively.In terms of specificity,the mean specificity of the random forest model,the support vector machine model and the neural network model are 0.61,0.21 and 0.33.According to the results of performance evaluation index,the stability and generalization ability of the random forest model are better than the support vector machine model and neural network model.Therefore,the random forest algorithm was selected to construct a forecasting model of S.avenae in eastern Shandong Province and to analyze the association between the numbers of S.avenae and environmental factors.Lady beetles,sunshine hours,and average temperature were most associated with S.avenae occurrence,with average reductions in Gini values of 12.56,7.65,and 5.32,respectively.The daily precipitation,average relative humidity,maximum wind speed,and average pressure have little effect on the occurrence of S.avenae,with the average reductions in Gini values were 1.43,1.12,1.05,and 0.89,respectively.Further research found that when the number of lady beetles(Coccinella septempunctata L.and Harmonia axyridis(Pallas))were more than 1 per square meter in wheat field,sunshine hours were more than 10 hours,and the temperature was higher than 18 ?,the number of S.avenae per hundred plants were more than 500.