Forecasting the occurrence of wheat stripe rust disease using time series study of Sentinel-2 satellite images: A case study of Ningyang, China.

Wheat stripe rust has a severe effect on the yield and quality of wheat. An effective forecasting method is essential for food security. In this study, firstly, optimal vegetation indices related to belt rust stress are extracted as candidate features for disease prediction from 2-Sentinel time series images. Then, optimal VI combinations are selected using sequential forward selection (SFS). Finally, the occurrence of wheat stripe rust in different time intervals is predicted using Support Vector Machine (SVM) method. The results of the selected traits show that, before the binding period, the optimal VIs are related to biomass, pigment and moisture of wheat. After the connection period, the red-edge VIs related to the health status of the product play an important role. The overall accuracy and kappa coefficient of the prediction model, which is based on SVM, is generally higher than K-nearest neighbor (KNN) and back-propagation neural network (BPNN) methods. SVM method is more suitable for predicting time series of wheat rust. The model achieved accuracy based on optimal VI combinations and the SVM increased over time. The highest accuracy was 86.2%. These results show that the prediction model can provide guidance and suggestions for early disease prevention in the study area, and this method combines Sentinel-2 time series images and SVM, which can be used to predict wheat stripe rust.