Calibration Model for Purple Air monitored PM2.5 across the United States

Area of Science:

• Environmental Science
• Air Quality Monitoring
• Data Science

Background:

• Accurate monitoring of fine particulate matter (PM2.5) is essential for assessing air quality and public health.
• Purple Air (PA) sensors offer low-cost, real-time PM2.5 data but face accuracy concerns due to biases from environmental factors and sensor drift.

Purpose of the Study:

• To develop and implement a scalable, operational calibration model for US-based Purple Air sensors.
• To enhance the reliability of PM2.5 data from PA sensors without external meteorological data.
• To generate multiple calibrated PM2.5 products catering to diverse end-user needs.

Main Methods:

• A random forest regression model was developed using PA-reported parameters (PM1, PM10, PM2.5, temperature, pressure, humidity) from 2019-2021.
• Calibration products were generated using varying spatial buffer distances around EPA monitoring sites.
• The model was evaluated on 2022 data, assessing performance using the Pearson correlation coefficient.

Main Results:

• The calibrated PA estimates showed strong agreement (Pearson correlation coefficient 0.90-0.96) with EPA reference PM2.5 measurements.
• High accuracy was maintained across different buffer zones (500 m to 5,000 m).
• The model demonstrated strong predictive performance without relying on complex deep learning architectures.

Conclusions:

• The developed PA-only calibration framework provides a practical, transparent, and reproducible solution for improving PM2.5 data reliability.
• This approach is particularly valuable for regions with limited access to regulatory-grade air quality infrastructure.
• The calibrated products support various applications, from local air quality assessments to national epidemiological studies.