Predicting plant disease epidemics using boosted regression trees.

Area of Science:

• Agricultural science
• Plant pathology
• Statistical modeling

Background:

• Plant epidemics, such as Fusarium head blight (FHB) in wheat, are frequently linked to weather patterns.
• Identifying precise weather predictors for epidemic prediction models remains a significant challenge.
• Previous research utilized scalar-on-function regression to model FHB epidemics based on weather time series.

Purpose of the Study:

• To reproduce and evaluate the findings of Shah et al. regarding Fusarium head blight (FHB) prediction in wheat.
• To compare the performance of boosted regression trees against previously established scalar-on-function regression models for FHB epidemic prediction.
• To assess the classification accuracy and statistical performance of boosted regression trees using a 140-day weather time series relative to anthesis.

Main Methods:

• Utilized a dataset and modeling framework previously employed by Shah et al.
• Applied boosted regression trees (BRT) as an alternative statistical approach.
• Modeled a binary outcome (FHB epidemic or non-epidemic) using daily weather time series data.

Main Results:

• Boosted regression trees demonstrated strong classification accuracy in predicting FHB epidemics.
• The model statistics achieved with boosted regression trees were found to be highly favorable.
• The performance of boosted regression trees was comparable or superior to the scalar-on-function regression models previously reported.

Conclusions:

• Boosted regression trees provide a robust and accurate method for predicting wheat Fusarium head blight (FHB) epidemics.
• This approach offers a valuable alternative for modeling plant disease epidemics based on weather data.
• The findings suggest that machine learning techniques like BRT can effectively identify complex weather-related predictors for agricultural disease forecasting.