Monitoring and Predicting Water Stress in Coffee Plants via Environmental Sensing and Plant Electrophysiology

Increasing climatic variability is intensifying water stress risks in coffee-producing regions, highlighting the need for efficient systems to monitor plant water status and optimize irrigation. The ConSenso project aims to develop innovative plant-based sensing technologies capable of detecting early physiological stress in coffee plantations to reduce water use and maintain crop health. This thesis investigates the integration of plant physiological signals with environmental measurements to improve the monitoring of water availability in coffee agroecosystems. The study was conducted in a Coffea arabica plantation in the Mbeya region of Tanzania, where a multi-sensor monitoring infrastructure was deployed. The system combines meteorological, soil, stem growth and resitivity data, collected continuously from September 2023 to September 2024. After preprocessing an exploratory analysis was performed to examine relationships among variables. A predictive modeling framework was then developed to estimate soil moisture dynamics using environmental and plant-based measurements: soil moisture was used as a proxy potential stress conditions within the soil–plant–atmosphere system. Three models were compared: Linear Regression, Random Forest, and Gradient Boosting. Machine learning approaches significantly outperformed the linear baseline model, emphasizing the non-linear interactions among variables. The inclusion of plant-based signals, including dendrometer measurements and electrical resistivity, substantially improved predictive performance. The optimized Gradient Boosting model achieved the best results, explaining about 95% of the observed variability in soil moisture (R² =0.95) with a root mean squared error of approximately 4.05. These findings demonstrate the potential of integrating plant electrical sensing technologies with environmental monitoring systems to improve soil water prediction and support data-driven irrigation management in coffee agroecosystems.