A0367
Title: Spatial filtering for unified calibration of air pollution data from multiple low-cost sensor networks
Authors: Claire Heffernan - Merck (United States) [presenting]
Abstract: Low-cost air pollution sensors are increasingly being deployed worldwide, creating networks that provide information on local variability within a region, but these sensors have considerable measurement error. In some cities, including Baltimore, Maryland, there are multiple low-cost networks covering the same area, providing several sources of information about air quality. While there are many available methods to calibrate data from one low-cost network, separate calibration of each network leads to conflicting predictions of air quality. A unified Bayesian spatial filtering model is developed that combines data from multiple low-cost networks as well as any available reference devices in the region to provide unified predictions at any location within the region. The method allows for network-specific calibration equations as biases and noise levels of low-cost sensor networks depend on the type of sensor used since the measurement error varies across networks. Also, the method guards against potential preferential sampling of some of the networks, providing better predictions and narrower confidence intervals compared to calibrating each network individually. The method is fit to PM$_{2.5}$ data in Baltimore in June and July 2023. The approach can be used to calibrate low-cost air pollution sensor data from multiple sensor types in Baltimore going forward.
