Robust statistical calibration and characterization of portable low-cost air quality monitoring sensors to quantify real-time O3 and NO2 concentrations in diverse environments

Abstract

Low-cost sensors offer an attractive solution to the challenge of establishing affordable and dense spatio-temporal air quality monitoring networks with greater mobility and lower maintenance costs. These low-cost sensors offer reasonably consistent measurements but require infield calibration to improve agreement with regulatory instruments. In this paper, we report the results of a deployment and calibration study on a network of six air quality monitoring devices built using the Alphasense O3 (OX-B431) and NO2 (NO2-B43F) electrochemical gas sensors. The sensors were deployed in two phases over a period of 3 months at sites situated within two megacities with diverse geographical, meteorological and air quality parameters. A unique feature of our deployment is a swap-out experiment wherein three of these sensors were relocated to different sites in the two phases. This gives us a unique opportunity to study the effect of seasonal, as well as geographical, variations on calibration performance. We report an extensive study of more than a dozen parametric and non-parametric calibration algorithms. We propose a novel local non-parametric calibration algorithm based on metric learning that offers, across deployment sites and phases, an R2 coefficient of up to 0.923 with respect to reference values for O3 calibration and up to 0.819 for NO2 calibration. This represents a 4-20 percentage point increase in terms of R2 values offered by classical non-parametric methods. We also offer a critical analysis of the effect of various data preparation and model design choices on calibration performance. The key recommendations emerging out of this study include (1) incorporating ambient relative humidity and temperature into calibration models; (2) assessing the relative importance of various features with respect to the calibration task at hand, by using an appropriate feature-weighing or metric-learning technique; (3) using local calibration techniques such as k nearest neighbors (KNN); (4) performing temporal smoothing over raw time series data but being careful not to do so too aggressively; and (5) making all efforts to ensure that data with enough diversity are demonstrated in the calibration algorithm while training to ensure good generalization. These results offer insights into the strengths and limitations of these sensors and offer an encouraging opportunity to use them to supplement and densify compliance regulatory monitoring networks.