Research On Monitoring And Prediction Of Wheat Fusarium Head Blight At The Regional Scale Based On Multi-source Data

The occurrence and spread of Fusarium head blight(FHB)pose a serious threat to wheat yield and quality.Timely,accurate,and effective spatial monitoring and prediction of FHB occurrence at the regional scale can guide the precise and rational application of pesticides,thereby achieving timely disease prevention and control,resource conservation,environmental protection,and improved wheat quality and yield.In this paper,wheat FHB was taken as the research object,and a regional-scale monitoring and prediction model for FHB was constructed using satellite remote sensing data,meteorological data,and field investigation data.The specific research content and results are as follows:(1)Extracting wheat planting area in the study area.The extraction of wheat planting areas is an important background information for monitoring and predicting wheat Fusarium head blight.In this paper,based on Sentinel-2 remote sensing data,the maximum likelihood method,support vector machine,and random forest were used to classify different land cover types(wheat,non-wheat vegetation,water bodies,bare land,buildings,and roads)and extract the wheat planting area.It was verified that the random forest algorithm yielded the best results with a classification accuracy of91.8%.This provides a spatial data foundation for subsequent remote sensing monitoring and prediction of diseases.(2)Establishment of a regional-scale monitoring model for Fusarium head blight severity in wheat.In this study,a library of monitoring characteristic factors for wheat Fusarium head blight was constructed based on multi-temporal meteorological and remote sensing data.The Relief FPearson algorithm was used to determine the optimal single-temporal spectral feature set,optimal two-temporal spectral feature set,and optimal meteorological feature set.Subsequently,wheat Fusarium head blight monitoring models were constructed using the adaptive boosting algorithm,support vector machine,and random forest classifier algorithm,with the optimal spectral feature set,optimal two-temporal spectral feature set,and optimal meteorological feature set as input variables,respectively.The results showed that the monitoring model constructed using the combination of the spectral feature set and meteorological feature set was superior to the model constructed using only the spectral feature set.The monitoring model constructed using the optimal two-temporal spectral feature set and meteorological feature set as input variables,and the random forest algorithm as the classifier,had the best performance,with an overall accuracy of 82.4% and a Kappa coefficient of 0.734.This study provides new ideas and methods for the effective monitoring of wheat Fusarium head blight at the regional scale.(3)Establishment of SEIR-FHB predictive model for wheat Fusarium head blight(FHB)incidence based on multi-source spatiotemporal information.Coupling the classic epidemiological model SEIR with meteorological factors and spectral characteristics that influence and characterize the occurrence and development of FHB in wheat,the SEIR-FHB predictive model was constructed to achieve dynamic predictions of FHB incidence in the study area.The predictive accuracy of the model was R~2=0.79 and RMSE=0.098.Compared with the FHB predictive models established by the least squares method and ridge regression method,the SEIR-FHB model had better predictive performance,with R~2 higher by 0.11 and 0.07,and RMSE lower by 0.069 and 0.039,respectively.Furthermore,to further validate the predictive performance of the SEIR-FHB model,actual incidence rate data in the study area were used for verification,and the results showed that the predictive results of the SEIR-FHB model were satisfactory and consistent with the actual data.Finally,the SEIR-FHB model constructed in this study provided spatiotemporal dynamic wheat FHB early warning information.