Regional prediction of Fusarium head blight occurrence in wheat with remote sensing based Susceptible-Exposed-Infectious-Removed model

• Published in: International journal of applied earth observation and geoinformation 2022-11, Vol.114, p.103043, Article 103043
• Main Authors: Xiao, Yingxin, Dong, Yingying, Huang, Wenjiang, Liu, Linyi
• Format: Article
• Language: English
• Subjects: Disease progress curves; FHB epidemiology; Fusarium head blight in wheat; Polycyclic diseases; Regional prediction; Remote sensing; Susceptible-Exposed-Infectious-Removed model
• ISSN: 1569-8432; 1872-826X
• DOI: 10.1016/j.jag.2022.103043

•Fusarium head blight (FHB) is a major fungal disease affecting wheat production.•Regional disease prediction using mechanistic models is difficult.•A remote sensing based model for regional prediction of FHB in wheat is proposed.•Optimized soil adjusted vegetation index (OSAVI) was used for model initialization.•Susceptible-Exposed-Infectious-Removed (SEIR) model was calibrated and verified. Fusarium head blight (FHB) is one of the major fungal diseases affecting wheat production worldwide, influencing kernel development and producing poisonous mycotoxins. Mechanistic models have been extensively used for plant disease simulation; however, regional disease prediction using these models is difficult because they simplify the heterogeneous plant growth conditions. Herein, we present a remote sensing based Susceptible-Exposed-Infectious-Removed (SEIR) model for regional prediction of FHB occurrence in wheat. Plant properties that are key to the development of FHB are extracted from remote sensing data or data products to initialize or drive the model. Fractional vegetation cover products, time-series curves from satellite images, and vegetation indices were used to indicate plant density, phenology, and vegetation vigor. We applied our model to a plain region in China that suffers greatly from FHB annually. The SEIR model was parameterized by incorporating remote sensing data products, and then calibrated and verified for regional FHB prediction. The model was trained and evaluated by comparing the results of its prediction of FHB incidence to field observations during the susceptible period for wheat; satisfactory results were observed with a correlation coefficient of 0.804, root mean-square error of 0.131, classification accuracy of 0.860, and missed detection rate of 0.035 when the model was initialized with the Optimized Soil Adjusted Vegetation Index (OSAVI). The disease progress curves furnished by our model display an S-shape—a characteristic of polycyclic diseases—which matches the wheat FHB epidemiology. These results indicate that our remote sensing-based SEIR model is promising for the regional prediction of FHB occurrence in wheat.