E-nose and AI system for early detection of plant water stress

Overview

This solution uses an electronic nose (E-Nose) combined with artificial intelligence to detect water stress in soybean plants before visible symptoms appear. Water stress triggers physiological and biochemical changes in plants, including stomatal closure, reduced photosynthesis, and altered respiration, which affect the volatile gases emitted by the plant canopy. By capturing and analyzing these gas signatures, the system can non-invasively assess the plant's water status and phenotypic condition.

The technology is particularly relevant for soybean production, where timely irrigation management is critical for yield optimization. Early detection of water stress enables farmers and agronomists to apply precision irrigation, reducing water waste and improving crop outcomes. Beyond soybean, the approach has potential applicability to other crops where transpiration-related gas emissions can serve as indicators of plant health.

Technical specifications

How it works:

• Gas samples are extracted from sealed containers housing soybean plants
• An E-Nose sensor array detects volatile compounds in the air surrounding the plant canopy
• Environmental sensors continuously measure relative humidity (%), temperature (°C), and CO2 (ppm) every five minutes
• Machine learning algorithms, including decision trees, classify the plant's water stress status based on gas signatures

Key features:

• Non-invasive phenotyping approach for water stress detection
• Decision tree classifier achieving 92.0% accuracy in distinguishing irrigated from water-stressed plants
• Based on 500 gas samples collected across six experimental trials with paired soybean plants over 30-day cycles
• Integration of environmental monitoring (RH, temperature, CO2) to contextualize gas readings
• Designed for compatibility with future machine learning and data mining techniques for severity grading

Experimental setup:

• Six experiments conducted with six pairs of soybean plants (BRs type)
• 10 days of normal irrigation followed by 20 days of water stress per cycle
• Soil surface covered to eliminate evaporation interference

Technology readiness level

The technology is at an early-to-mid stage of development. Preliminary validation has been completed with 500 gas samples yielding 92% classification accuracy between irrigated and water-stressed conditions. The research team at Embrapa Soja plans a 1.5-year follow-up study to advance the system toward controlled-condition measurements, early stress detection, and severity classification.

Future development requires new plant containers, a portable E-Nose, additional sensors, software, computing infrastructure, and dedicated personnel. The team is actively seeking partnership with a private company to align the technology with market needs and guide commercial development. Current readiness is suitable for collaborative research, pilot validation, and co-development toward a deployable precision agriculture tool.