Low-cost Sensor Array Devices as a Method for Reliable Assessment of Exposure to Traffic-related Air Pollution

摘要

The exposure to air pollutant mixtures is a well-known risk factor for inducing and increasing the severity of diseases. Combining low-cost sensors in arrays holds great potential for real-time monitoring of pollutant exposure because of versatility and aptitude for tracking multi-pollutant exposure. While many low-cost air pollution monitoring devices have been proposed, several underexplored opportunities remain, including sensor-derived pollution indices, source analysis and exposure assessment. A thorough investigation of different low-cost commercial gas and particulate matter sensors from 5 manufacturers has been conducted and best-performing sensors were identified. A device for monitoring air quality has been developed and tested. Each device consists of an array of commercially available metal oxide semiconductor for monitoring NOx and 03, CO, C02 and optical sensors for monitoring PM2.5. Level of pollutant exposure has been characterized at different locations in Toronto over 3 different campaigns between 2013 and 2016. These deployments allowed long-term sensor performance to be evaluated under different meteorological conditions as well as different ranges of pollutant concentrations. Analysis of a large range of gas sensors revealed several key challenges, including high intra-sensor variability, interference from temperature and nonlinearity. Air quality health index estimation from sensor readings was successfully demonstrated. Three aspects of device reproducibility were evaluated: drift over time, impact of interferences and impact of site-specific mixtures. Three categories of approaches for improving sensor accuracy and reproducibility were tested: nonlinear calibration models, variable transformations and training data selection. Model reproducibility, ability to adjust for multiple combinations of interferences and ability to resolve sites was improved when devices were calibrated at multiple sites. Analysis showed that both short-term and long-term temporal patterns could be resolved and compared at different sites. Background subtraction helped further emphasize the differences and rank sites in terms traffic-related pollution.