多源数据小麦条锈病预测研究

小麦条锈病是导致小麦大规模减产的气传性病害，其传播扩散过程受多种因素影响，常用的作物病害气象预测模型难以准确模拟。为实现小麦条锈病发病率的精准预测，提出一种基于气象和遥感数据建立的SEIR-StripeRust动态预测模型。以甘肃省陇南地区为研究区，首先基于气象数据和MODIS遥感数据分别构建气象因子和植被指数，然后与发病率进行相关性分析筛选敏感因子并耦合基本感染率，进而建立SEIR-StripeRust模型，最后采用后向传播神经网络（BPNN）、支持向量回归（SVR）和多元线性回归（MLR）模型对比验证SEIR-StripeRust模型的有效性。结果表明：平均气温、相对湿度和归一化植被指数与小麦条锈病发病率显著相关，其建立的SEIR-StripeRust模型预测精度最高，决定系数R² 达到0.79，均方根误差RMSE为0.10，平均绝对误差MAE为0.09，均优于相同特征变量下的BPNN、SVR和MLR模型。研究结果表明SEIR-StripeRust模型能够有效预测小麦条锈病发病率，并为县域尺度的小麦条锈病预测和精确防控提供技术支持。

Research on Prediction of Wheat Stripe Rust with Multi-source Data

Wheat stripe rust is an air-borne disease that leads to large reduction in wheat production. The spread process is affected by many factors. Common crop diseases meteorological prediction models are difficult to simulate wheat stripe rust incidence accurately. In order to obtain accurate prediction of wheat stripe rust incidence， a Suscept-Exposed-Infectious-Removed StripeRust （SEIR-StripeRust） dynamic prediction model was constructed based on meteorological and remote sensing data. This paper chose the Longnan area of Gansu Province as a study area. First， meteorological factors and vegetation indexes were constructed based on meteorological data and MODIS data， respectively. Then， the above features were screened by correlation analysis to identify the sensitive factors. A new incidence prediction model named SEIR-StripeRust was constructed， coupled with the sensitive factors. Finally， compared the accuracy of SEIR-StripeRust model with used Back Propagation Neural Network （BPNN）， Support Vector Regression （SVR） and Multiple Linear Regression （MLR）. The results showed that the average temperature， relative humidity and normalized difference vegetation index were significantly correlated with the incidence of wheat stripe rust. The SEIR-StripeRust model constructed by the above three sensitive factors had the highest prediction accuracy， the coefficient of determination （R² ） was 0.79， the Root Mean Square Error （RMSE） was 0.10， and the Mean Absolute Error （MAE） was 0.09， which were higher than the BPNN， SVR and MLR models under the same characteristic variables. The results showed that the SEIR-StripeRust model can effectively predict the incidence of wheat stripe rust and provide technical support for wheat stripe rust prediction and accurate prevention at county scale.