Detecting Plant VOCs With Indoor Air Quality Sensors

Abstract

Buildings need practical ways to monitor indoor air quality (IAQ) beyond aggregate TVOC readings. We show that low-cost commercial VOC sensors, coupled with machine learning, can recover compound-specific information from plant-emitted terpenes, enabling practical, real-time bioindication in buildings. In an office testbed, we exposed sensors to 16 terpenes and trained random forest, support vector machine, and XGBoost models on time series features. The models detected “any terpene versus background” at 97%–100% accuracy, identified “plants versus background” at ~100%, and discriminated among individual compounds with accuracies up to 96%. Feature importance emphasized temporal dynamics (e.g., autocorrelation lags and entropy measures) rather than static peaks, highlighting the value of sequence information for commodity hardware. Complementary experiments with living basil plants showed reproducible VOC profiles and stress-induced bursts of ~70–100 ppb, confirming in situ feasibility. A placement analysis across 13 locations indicated that the HVAC return-air duct provides the most actionable, room-integrated signal for deployment, balancing accuracy and coverage. Together, these results establish a pathway from TVOC to compound-aware IAQ using sensors already common in smart buildings, with immediate applications to exposure triage and demand-controlled ventilation, and a foundation for plant-integrated environmental monitoring.