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

Accurate monitoring of PM2.5 is vital for understanding air quality and public health impacts. Purple Air sensors (PA) have filled in PM2.5 gaps spatially and temporarily, and are popular for their low cost and real-time monitoring capabilities, but are often questioned for their accuracy. Analysis revealed that PA sensors exhibit systematic biases in PM2.5 measurements, with discrepancies influenced by factors such as humidity, temperature, and sensor drift. To address this, we develop and implement a scalable, operational calibration model for the United States based solely on the PA-reported parameters, including particulate matter (PM1, PM10, PM2.5), temperature, pressure, and relative humidity, without any reliance on external meteorological fields or ancillary datasets for operational stability. To address diverse end-user requirements, we generate multiple calibrated PM2.5 products using varying spatial buffer distances around Environmental Protection Agency (EPA) monitoring sites. A relatively simple, interpretable random forest regression approach is employed, avoiding computationally intensive or opaque deep neural network architectures while maintaining strong predictive performance. The model was built using data from 2019-2021 and then evaluated on 2022 data. Its performance, assessed using the Pearson correlation coefficient, shows strong agreement (0.90-0.96) between calibrated PA estimates and EPA reference PM2.5 measurements across buffer zones ranging from 500 m to 5,000 m. By prioritizing transparency, operational simplicity, and reproducibility, this framework provides a practical and user-oriented solution for enhancing the reliability of PA PM2.5 data, particularly in regions lacking access to regulatory-grade infrastructure. The resulting pipeline of PA-only, EPA-referenced calibration products supports applications ranging from localized air quality assessment to national-scale exposure and epidemiological studies.