Weather-Based Logistic Regression Models for Predicting Wheat Head Blast Epidemics

Area of Science:

• Agricultural Science
• Plant Pathology
• Computational Biology

Background:

• Wheat head blast poses a significant threat to wheat production in Brazil's Cerrado region.
• Accurate prediction of disease epidemics is crucial for effective crop management and minimizing yield losses.

Purpose of the Study:

• To develop and validate empirical models for predicting wheat head blast epidemics in Brazil.
• To identify key weather variables and time windows predictive of disease outbreaks.

Main Methods:

• Collected data from 143 wheat head blast epidemics across multiple Brazilian sites (2012-2020).
• Utilized daily weather data from NASA POWER and wheat heading date (WHD) to create 36 potential predictors.
• Employed logistic regression with LASSO and best subset selection for model development, validated using leave-one-out cross-validation (LOOCV).

Main Results:

• Models with 2-5 predictors achieved high performance: accuracy (0.80-0.85), sensitivity (0.80-0.91), specificity (0.72-0.86), and AUC (0.89-0.91).
• LOOCV accuracy ranged from 0.76-0.81.
• The final model incorporated preheading temperature/humidity and postheading precipitation, accurately predicting outbreaks in a 24-year series.

Conclusions:

• Developed a robust predictive model for wheat head blast outbreaks suitable for tropical and subtropical climates.
• The model's accuracy in predicting historical epidemics validates its potential for practical application in disease management.
• This tool can significantly aid farmers and researchers in mitigating the impact of wheat head blast.