A mechanistic model integrating multi-source data for operational prediction of rice sheath blight at regional scale

Rice sheath blight (RSB) is a destructive disease that causes significant yield losses globally. Accurate regional prediction is critical for sustainable disease management; however, existing models suffer from poor generalizability and limited operational applicability. This study aims to develop a mechanistic, data-driven prediction model integrating epidemiological and environmental factors to improve operational prediction of RSB.

We developed the susceptible-exposed-infected-removed model for rice disease (SEIR-RICEDIS) by coupling an SEIR epidemiological framework with three driving modules: temperature, precipitation, and annual peak value of disease incidence. The model was parameterized and evaluated using field survey records from 2010 to 2015, meteorological observations, and satellite remote sensing data from the Middle-Lower Yangtze Plain. Model parameters were optimized using a genetic algorithm. Across two validation scales, the model achieved an R² of 0.63-0.72 and a mean root mean square error of 6.68-10.36 for disease incidence prediction. Disease-control recommendations derived from the multi-phase relative area under the disease progress curve achieved an accuracy of 80-84%.

The developed SEIR-RICEDIS model effectively simulated the spatio-temporal epidemic dynamics of RSB. It provides a robust, mechanism-based framework for operational disease forecasting at a regional scale, facilitating decision-making in disease control. © 2026 Society of Chemical Industry.