Work

An Adaptive Dual-Stage Learning System to Predict the Water Needs of Tomato Plants Exploiting Bioristor Data

Year: 2026
Type: conference-paper
Abstract: The increasing pressure on water resources requires intelligent and scientifically informed solutions for developing adaptive precision irrigation in agriculture. Herein, we propose a dual-stage learning system that captures complex, non-linear, physiological dynamics from edge data, measured through local in-vivo biosensors, with the final aim of supporting autonomous irrigation decision-making. The system exploits measurements from the so-called “Bioristor”, an innovative plant-embedded biosensor that continuously monitors internal ionic dynamics in the xylem sap, providing early indicators of water stress in the plants. The proposed architecture decomposes the irrigation problem into two sequential tasks: i) a binary classification stage on whether irrigation is required and ii) a regression stage estimating the amount of water to be delivered day-by-day. This formulation improves robustness under class imbalance and avoids issues related to zero-inflated targets. The system is evaluated on real-world data collected from an open-field tomato crop. The experimental results show that the proposed approach achieves accurate irrigation decisions and precise water amount estimation, with lightweight neural models emerging as an effective trade-off solution for edge-oriented deployment.
Language: English
Subfield: Plant Science
Open Access status: hybrid
OpenAlex ID: W7225751679
FWCI:
Cites: 15
Cited by:
Related to: 10