Climate and disease: tackling coffee brown-eye spot with advanced forecasting models

Area of Science:

• Agricultural Science
• Plant Pathology
• Data Science

Background:

• Climate significantly impacts host-pathogen interactions, crucial in agriculture.
• Fungal diseases like Cercospora coffeicola (brown-eye spot) severely affect coffee yields.
• Accurate disease forecasting is vital for effective crop management.

Purpose of the Study:

• To develop and evaluate machine learning models for forecasting coffee brown-eye spot.
• To predict disease incidence at least one week prior to occurrence.
• To spatially project disease severity across key Brazilian coffee-producing regions.

Main Methods:

• Utilized agrometeorological data (temperature, humidity, rainfall, etc.) from multiple locations.
• Employed six machine learning models: KNN, MLP, SVM, RF, XGBoost, and GradBOOSTING.
• Incorporated Penman-Monteith and Thornthwaite and Mather methods for water balance calculations.

Main Results:

• XGBoost model excelled in high-yielding conditions (R²=0.77), while SVM performed better in low-yielding scenarios (precision=0.76).
• Disease incidence varied significantly between high- and low-yield conditions and across regions.
• Prediction accuracy was influenced by the biennial coffee production cycle.

Conclusions:

• Agrometeorological variables and machine learning models effectively predicted brown-eye spot incidence with a 7-day lead time.
• These predictive models offer valuable tools for managing coffee brown-eye spot.
• The study highlights the potential of data-driven approaches in agricultural disease management.