FILTER: Framework for Improving Low-Cost Sensor Network Data for Air Quality Monitoring

Over the last decade, low-cost sensors (LCSs) have improved air quality monitoring by enabling widespread, community-driven data collection, particularly in regions with limited resources. Although these LCSs have increased public engagement and enriched datasets for understanding pollution dynamics, challenges related to data quality, standardization, and interoperability have hindered their full integration into regulatory frameworks and large-scale environmental monitoring (Barkjohn et al., 2024; Carotenuto et al., 2023). The lack of consistent Quality Control (QC) processes and correction methodologies limits the reliability of LCS-derived data for applications such as public health assessments, modeling, and policymaking.
To address these issues, we introduce FILTER (Framework for Improving Low-cost Technology Effectiveness and Reliability), a scalable, multi-level QC, and correction framework designed to improve the reliability of PM2.5 data from citizen-operated LCS networks. FILTER employs spatial “correlation” and “similarity” QC tests, allowing for in-situ correction of LCS data in different environments. The FILTER framework’s effectiveness is validated using large-scale European data from the sensor.community and PurpleAir networks, two of Europe’s largest citizen-driven air quality networks, covering the period 2018 – 2023. The final dataset, including 521,115,762 hourly PM2.5 measurements from 37,085 locations, was categorized into “high quality,” “good quality,” and “other quality” groups. At the raw data stage, applying QC steps through the spatial similarity level results in a ~50.3% decrease in median RMSE (from 7.61 to 3.78 µg m⁻³) across 483 LCSs. For corrected data, applying the same QC steps reduces the median RMSE by ~49.5% (from 7.59 to 3.83 µg m⁻³) across 456 LCSs. These enhancements enable LCS data to be more effectively integrated into scientific research, regulatory datasets, and policy development.
FILTER demonstrates several key advantages: independence from sensor-specific designs, geospatial scalability, adaptability to real-time processing, and applicability to other pollutants with spatial patterns similar to PM2.5. However, its utility is constrained for pollutants like NO₂, which exhibit hyper-local variability, or scenarios requiring sub-hour temporal resolution. FILTER also demonstrates the potential of on-site calibration techniques that do not require co-location to generate accurate data from LCS networks. This is particularly important for developing large LCS networks that can contribute to both science and policy, especially in light of the new European Air Quality Directive. 

References
Barkjohn, K. K., Clements, A., Mocka, C., Barrette, C., Bittner, A., Champion, W., et al. (2024). Air Quality Sensor Experts Convene: Current Quality Assurance Considerations for Credible Data. ACS ES&T Air. 
Carotenuto, F., Bisignano, A., Brilli, L., Gualtieri, G., & Giovannini, L. (2023). Low‐cost air quality monitoring networks for long‐term field campaigns: A review. Meteorological Applications, 30(6), e2161. 

We acknowledge funding for CitiObs project from the European Union’s Horizon Europe research and innovation programme under grant agreement No.101086421. We also acknowledge the contributions of sensor.community (https://sensor.community/en/, accessed October 2024) and PurpleAir (https://www2.purpleair.com/, accessed October 2024) sensor networks where the original sensor data come from, as well as the citizens who provided the low-cost sensor data.