Surface Pressure as a Nontraditional Predictor of Coffee Leaf Rust (H. vastatrix): A Machine Learning Ensemble Analysis from Costa Rican Climatic Data

Coffee Leaf Rust (CLR) remains an extremely damaging fungus disease to coffee crops globally, caused by the fungus Hemileia vastatrix . For Costa Rica, a country with major reliance on coffee exports, the CLR epidemic of 2012–2013 brought about national emergencies which showcased how coffee growing nations are prone to the impacts of climate related disease dynamics. Previous studies on CLR outbreaks have focused on the temperature, precipitation, and relative humidity, but atmospheric surface pressure remains mostly unexplored, and is a more holistic factor combining both elevation/weather variation and climate dynamics. Using a Voting Regressor ensemble model with CatBoost, XGBoost, and Gradient Boosting algorithms, this study aims to examine surface pressure as a nontraditional predictor of CLR outbreaks. The model utilized publicly accessible climate data from 2014 to 2019 and CLR incidence in major coffee producing regions of Costa Rica. To evaluate the importance and contribution of each variable, surface pressure as well as traditional climate variables were assessed. As shown in the results, surface pressure emerged as a significant and a major predictor of CLR incidence, even more than temperature and precipitation. Particularly, model analysis interpreting a simple linear regression model showed that CLR rose by 1.26% for every 10 hPa rise in mean surface pressure, indicating a moderately strong positive correlation, related mostly with altitude. While this estimation makes a simplification of reality, our ensemble based Mutual Information analysis (MI = 0.1817) confirms that surface pressures influence on CLR does have a considerable impact and is non-linear, and on average, changes in surface pressure (both positive and negative depending on the region) of around 5–9 hPa indicated increases in CLR incidence. Specifically, the ensemble model learned to identify certain pressure thresholds, for example, values above 1005 hPa, where the rate of rust incidence increase was much greater at Turrialba and Coto Brus when compared to other locations within the study. In this case, the linear regression coefficient gives a simple, reasonable estimate of the biological model's logic, and the ensemble regression provides a fuller picture of its intricacy. This indicates the surface pressure not only captures elevation linked vulnerability, but also indicates some temporary meteorological conditions favorable for fungal growth. The broader implications of this research advocate for incorporating surface pressure into CLR real-time forecasting and CLR early warning systems, improving the effectiveness and flexibility of disease management planning. Thus, this study helps create efficient climate change adaptation tools for coffee farmers by pinpointing surface pressure as a dependable and straightforward climate indicator.