Remote Calibrating a National-Wide Dense Low-Cost Sensor Network for Hyperlocal Air Surveillance.
TL;DR
This study proposes a scalable remote calibration framework that eliminates physical colocation by integrating metropolis-specific monthly updates with a spatiotemporal pairing algorithm for representative training data selection and serves as a cornerstone in advancing the use of large-scale LCS networks.
Abstract
Emerging low-cost sensor (LCS) technologies offer a unique opportunity to achieve equitable, high-granularity air surveillance. Yet without proper calibration, their raw observations are often considered invalid for rigorous research and regulatory purposes. Common calibration practices require "case-by-case" colocation of low-cost sensors against "gold standard" reference stations, lacking scalability for decentralized, dense sensor networks. This study proposes a scalable remote calibration framework that eliminates physical colocation by integrating metropolis-specific monthly updates with a spatiotemporal pairing algorithm for representative training data selection. We validate this approach using over 33 million hourly observations from the largest U.S. low-cost sensor network across ten major metropolitan areas. The calibrated LCS observations exhibit good agreement with reference stations in all regions. We further observe that the spatial distribution of LCS varies significantly across racial/ethnic and income groups, with African American communities the least covered. The usefulness of calibrated LCS observations is demonstrated through exemplary applications of high-granularity land use regression and wildfire plume tracking. Our study serves as a cornerstone in advancing the use of large-scale LCS networks, from citizen science and public awareness to rigorous air quality research and policymaking.