AI-enhanced spatio-temporal volatile organic compound profile characterization integrated with GIS for environmental monitoring using Spectrometer-On-Card

Volatile organic compounds (VOCs) are air pollutants that harm health and contribute to climate change. This paper demonstrates how a portable VOC measuring device, combined with advanced signal processing, effectively characterizes VOC patterns. A novel methodology is introduced, leveraging deep learning to process temporal profiles from nine-channel nano-sensor technology. This is demonstrated through two use cases: VOC-pattern classification between indoor and outdoor environments, and VOC-pattern classification across different geographical settings—rural, urban, and industrial. For the first use case, the best RNN model achieves F-scores of 90.56% for 50-second window classification and 95.90% for exposure classification. In the second use case, it achieves F-scores of 81.76% and 86.53% for the same metrics. The best model is integrated into a GIS for detecting localized pollution hotspots. This low-cost, portable, calibration-free methodology has applications in industrial leak detection, plant health monitoring, and detection of human health issues.