Detecting Plant VOC Traces Using Indoor Air Quality Sensors

TL;DR

This study investigates using readily available indoor VOC sensors to detect and classify plant-emitted terpenes in real indoor environments. By combining three experiments with physics-based emission modeling and machine learning on time-series features, the authors show that commercial sensors can detect several terpenes and, with optimized sensor placement, distinguish between different VOC signatures and plant stress states. The work identifies key features (e.g., autocorrelation lag, permutation entropy, approximate entropy) that drive classification and demonstrates that live basil plant emissions can be detected and interpreted in a controlled setting. Overall, the research provides a foundational proof-of-concept for plant-informed sensing in smart buildings and points to practical avenues for sensor networks, stress monitoring, and future plant-based IAQ technologies.

Abstract

In the era of growing interest in healthy buildings and smart homes, the importance of sustainable, health conscious indoor environments is paramount. Smart tools, especially VOC sensors, are crucial for monitoring indoor air quality, yet interpreting signals from various VOC sources remains challenging. A promising approach involves understanding how indoor plants respond to environmental conditions. Plants produce terpenes, a type of VOC, when exposed to abiotic and biotic stressors - including pathogens, predators, light, and temperature - offering a novel pathway for monitoring indoor air quality. While prior work often relies on specialized laboratory sensors, our research leverages readily available commercial sensors to detect and classify plant emitted VOCs that signify changes in indoor conditions. We quantified the sensitivity of these sensors by measuring 16 terpenes in controlled experiments, then identified and tested the most promising terpenes in realistic environments. We also examined physics based models to map VOC responses but found them lacking for real world complexity. Consequently, we trained machine learning models to classify terpenes using commercial sensors and identified optimal sensor placement. To validate this approach, we analyzed emissions from a living basil plant, successfully detecting terpene output. Our findings establish a foundation for overcoming challenges in plant VOC detection, paving the way for advanced plant based sensors to enhance indoor environmental quality in future smart buildings.

Figures (8)

• Figure 1: Flowchart showing an overview of the Experiments
• Figure 2: This figure shows the layout of the room for the terpene experiment 1. Figure (a) is a representation of the room and Figure (b) is the actual deployment. 4 sensors were arranged on a table with two of them placed $75~cm$ away and the other two placed $125~cm$ away from the terpene/plant location respectively
• Figure 3: Schematic of sensor placement in Experiment 2 - Testing the VOC terpenes in the room using 13 spatially arranged sensors in the room. The sphere is the placement of the sample and the cubic shapes are the locations of sensors in the room including 6 sensors on the walls, 4 sensors on the table, 1 sensor on the cabinet, 1 sensor on the floor, and 2 sensors on the ceiling, attached to supply and return of the HVAC system. In the picture, sensors are defined with cubic boxes and their symbols from\ref{['tab.Sensor_summary']}
• Figure 4: Experiment 3 setup. The basil plant was placed in the center of the box in proximity to a sensor for 30 minutes of testing. The box had no air input or output
• Figure 5: Data collection by four sensors in Experiment 1 detecting D-limonene terpene signals across four consecutive experiments. Proximity to the terpene source correlates with higher maximum VOC trace points. Sensor 3 (the red line), positioned at a height of 75 cm, records the highest altitude among the sensors. Also, Awair Sensors 1 (purple line) and 4 (blue line) had the lowest picks