适用于我国的稻纵卷叶螟长期预测模型

Long-term models suitable in China for forecasting rice leaf folder Cnaphalocrocis medinalis

为给我国稻纵卷叶螟Cnaphalocrocis medinalis防治提供前期预警,使用R语言软件对我国15个省市区稻纵卷叶螟发生等级与全球海温场资料进行遥相关分析,绘制相关系数的时空间分布图,筛选出显著相关海温区作为预测因子,根据各省市区虫情数据组建回归模型+判别模型、BP神经网络模型和支持向量机(SVM)模型,比较3种模型的历史回检率和预测完全准确率。结果显示,3种模型对稻纵卷叶螟发生等级均有一定的预测能力,其中判别模型+回归模型效果最好,预检完全准确率可达到75.0%,BP神经网络模型次之,预检完全准确率为68.2%,SVM模型预测效果最差,预检完全准确率为54.5%。进一步分析建模所使用的50个预测因子的空间位置,在南印度洋和北大西洋确定3个预测指标,预检准确率为94.4%。通过海温场数据建立的我国15个省市区稻纵卷叶螟发生等级预测模型,适用于长期预测预报。判别模型+回归模型更适合在样本量少、预测因子相关性强的地区建模,而根据预测因子空间分布选择的预测指标进行定性预测准确率更高。

In order to guide the prevention and control of rice leaf folder Cnaphalocrocis medinalis,the R software was used to explore the teleconnection correlation between the occurrence degree of C.medinalis and the global sea surface temperature(SST)data in the provinces,municipalities and autonomous regions of China and draw the spatial-temporal distribution map of correlation coefficient.Then,the significant relevant SST zone was selected as the predictive factor according to the provinces.The long-term forecasting models,including discriminant model+regression model,BP neural network model and Support Vector Machine(SVM)model were established for 15 provinces,municipalities and autonomous regions,respectively.By comparing the historical regression rate and prediction accuracy of the models,it was found that the complete accuracy rate of forecast by discrimination or regression model reached 75.0%,followed by BP neural network model with 68.2%and SVM model with 54.5%.Further analysis of the spatial location of the 50 predictors used in the modeling,three predictive indicators in the South Indian Ocean and the North Atlantic had a prediction accuracy of 94.4%.The results indicated that the SST data could be used to establish long-term prediction model for the occurrence degree of rice leaf roller in China with a high accuracy.The discriminant model+regression model were more suitable for areas with fewer samples and strong correlation of predictors,and the prediction indexes obtained by spatial analysis of significant relevant SST zone showed higher prediction accuracy.