Repository logo
 

Low-cost embedded systems for community-driven ambient air quality monitoring

Date

2022

Authors

Wendt, Eric, author
Volckens, John, advisor
Pierce, Jeffrey, committee member
Jathar, Shantanu, committee member
Pasricha, Sudeep, committee member

Journal Title

Journal ISSN

Volume Title

Abstract

Fine particulate matter (PM2.5) air pollution is a leading cause of death, disease and environmental degradation worldwide. Existing PM2.5 measurement infrastructure provides broad PM2.5 sampling coverage, but due to high costs (>10,000 USD), these instruments are rarely broadly distributed at community-level scales. Low-cost sensors can be more practically deployed in spatial and temporal configurations that can fill the gaps left by more expensive monitors. Crowdsourcing low-cost sensors is a promising deployment strategy in which sensors are operated by interested community members. Prior work has demonstrated the potential of crowdsourced networks, but low-cost sensor technology remains ripe for improvement. Here we describe a body of work aimed toward bolstering the future of community-driven air quality monitoring through technological innovation. We first detail the development of the Aerosol Mass and Optical Depth (AMODv2) sampler, a low-cost monitor capable of unsupervised measurement of PM2.5 mass concentration and Aerosol Optical Depth (AOD), a measure of light extinction in the full atmospheric column due to airborne particles. We highlight key design features of the AMODv2 and demonstrate that its measurements are accurate relative to standard reference monitors. Second we describe a national crowdsourced network of AMODv2s, in which we leveraged the measurement capabilities of the AMODv2 in a network of university students to analyze the relationship between PM2.5 and AOD in the presence of wildfire smoke in the United States. Finally, we propose a cloud screening algorithm for AOD measurements using all-sky images and deep transfer learning. We found that our algorithm correctly screens over 95% of all-sky images for cloud contamination from a custom all-sky image data set. Taken as a whole, our work supports community-driven air pollution monitoring by advancing the tools and strategies communities need to better understand the air they breathe.

Description

Rights Access

Subject

citizen science
embedded systems
mechatronics
computer vision
air pollution
machine learning

Citation

Associated Publications