Speaker
Description
Ambient air pollution is among the foremost environmental determinants of
population health in sub-Saharan Africa, yet the density of regulatory monitoring
infrastructure remains severely inadequate for real-time community-level
surveillance. This paper presents AIrSynQ, an Artificial Intelligence of Things
(AIoT) platform for continuous, calibrated air quality monitoring developed as a
University of the Witwatersrand spin-out and supported by the South African
Technology Innovation Agency (TIA). The system integrates low-cost multi-parameter
sensor nodes measuring PM$_{2.5}$, PM$_{10}$, NOx, CO, and volatile organic
compounds (VOCs) with an edge-to-cloud data pipeline built on LoRaWAN and LTE-M
backhaul, Azure IoT Hub stream processing, and a machine-learning inference layer
performing real-time anomaly detection and pollutant forecasting. The South African
Community Air Quality Monitoring (SACAQM) network, built on the AIrSynQ platform,
currently operates more than 30 active nodes across the Gauteng province, with a
500-unit rollout in progress. Diurnal PM$_{2.5}$ measurements reveal systematic
exceedances of the WHO 24-hour guideline of 15~$\mu$g~m$^{-3}$ at informal
settlement nodes, with peak concentrations exceeding 70~$\mu$g~m$^{-3}$ during
morning and evening domestic combustion periods. A lightweight LSTM autoencoder
deployed at the cloud inference layer demonstrates a precision-recall trade-off
superior to static threshold alerting, enabling timely community and clinical
notifications. The platform architecture, deployment methodology, anomaly detection
performance, and public health implications are described. The AIrSynQ system
constitutes a replicable AIoT e-health infrastructure model for resource-constrained
urban environments across the Global South.
