晚稻稻瘟病BP神经网络分区预报

应用相关分析方法分析了浙江省19个县1988～1999年晚稻稻瘟病发病与有关环境因子的关系,筛选了8个气象因子用于晚稻稻瘟病发生程度长期预报。根据各预报因子与稻瘟病发病程度相关性,采用邻接二维图论聚类分析法,将19个点(县)划分为4个生态区。每个生态区内运用BP神经网络技术建立模型,并进行拟合和试报。1997～1999年试报验证,在划分稻瘟病生态区的基础上,应用BP神经网络模型对稻瘟病进行长期预测预报是可行的,3年试报成功率分别是78.95％、84.21％和78.95％。文中还对该方法与过去常用的预报方法的试报结果作了比较。

Zoning forecast for rice blast in late rice by using BP neural network

Climatic data from 19 counties in Zhejiang Province during 1998 - 1999 were used to investigate by correlation analyses, and eight factors were selected for forecasting severity of rice blast (Magnaporthe grisea) in late rice. Using two-dimension adjacency optimal-tree cluster, four ecological regions were plotted according to the relationships between climatic factors and severity of rice blast in 19 locations. A zoning long-term forecast model (ZLTFM) of rice blast in late rice in Zhejiang Province was derived from four ecological regions by using BP neural network technology. Validating results showed that accuracies of model in 19 counties from 1997 - 1999 were 78.95% , 84.21% and 78.95% , respectively. These demonstrated ZLTFM is feasible for forecasting severity of rice blast. A comparison between ZLTFM and statistic forecasting model was also conducted.