Predictive modeling of coffee leaf rust severity in Brazil under different yielding conditions

Abstract Coffee leaf rust (Hemileia vastatrix) is one of the most significant diseases affecting coffee crops, with considerable impacts on yield, particularly during high-production years. This study aimed to forecast coffee leaf rust severity in Brazil using machine learning (ML) models and to analyze the spatial and temporal variability of the disease under different yield conditions. The models evaluated included Extreme Gradient Boosting (XGB), Random Forest (RF), Multi-Layer Perceptron (MLP), Support Vector Machine (SVM), and Multiple Linear Regression (MLR), using open-access meteorological data from the NASA-POWER system as predictive variables. The results indicated that XGB showed strong predictive performance, with R² = 0.5973 in the test dataset for high-yielding conditions and R² = 0.5039 for low-yielding conditions, performing comparably to other models depending on the evaluation metric. The generated maps revealed a seasonal pattern in coffee leaf rust severity, with lower values during the rainy months (January to March) and higher severity between May and August, particularly in regions such as Minas Gerais, São Paulo, and Goiás. Additionally, high-yielding conditions were associated with higher disease severity throughout the year, suggesting increased plant susceptibility. The findings highlight the potential of machine learning approaches combined with open climatic data for large-scale and reproducible disease monitoring. Predictive modeling can support the identification of potential risk patterns and may contribute to decision-support strategies, including climate-based monitoring and more targeted phytosanitary management. The proposed framework provides a promising approach for analyzing coffee leaf rust dynamics and may be useful in other coffee-producing regions, although broader validation is still needed.