Predicting wheat powdery mildew epidemics in China using meteorological data and machine learning approaches

Area of Science:

• Agricultural Science
• Plant Pathology
• Computational Biology

Background:

• Accurate prediction of plant diseases like wheat powdery mildew (WPM) is crucial for effective agricultural management.
• This study focuses on developing predictive models for WPM severity and occurrence using meteorological data in China.
• Previous methods lacked the precision needed for large-scale WPM management.

Purpose of the Study:

• To develop and validate machine learning models for predicting wheat powdery mildew severity degree and occurrence area.
• To identify key meteorological factors influencing WPM development.
• To generate national-scale WPM severity distribution maps for improved disease management.

Main Methods:

• Trained and cross-validated six machine learning algorithms using 411 meteorological variables from 48 Chinese counties (1981-2021).
• Utilized K-Nearest Neighbor (KNN) for severity prediction and spatial interpolation models (IDW, ordinary kriging) for occurrence area mapping.
• Employed random forest for climate variable importance ranking and chi-squared/error reference methods for model validation.

Main Results:

• Support vector machine and KNN models demonstrated high performance in predicting WPM severity, particularly using data from the coldest month and wheat jointing-heading stages.
• Eight key meteorological predictors were identified, enhancing prediction accuracy.
• The IDW_4.0 model proved superior for generating nationwide WPM severity distribution maps (1990-2019).

Conclusions:

• Machine learning models effectively predict WPM severity and occurrence area at a national scale using meteorological data.
• Visualizing WPM severity spatial patterns aids in developing targeted and improved management strategies for China.
• The developed models offer a significant advancement in data-driven plant disease management.