Short-Term Machine-Learning Calibration of PID Sensors for Ambient VOC OH Reactivity

Area of Science:

• Environmental Science
• Analytical Chemistry
• Sensor Technology

Background:

• Photoionization detector (PID) sensors are cost-effective for monitoring volatile organic compounds (VOCs).
• PID sensor accuracy is limited by environmental factors (temperature, humidity) and sensor variability.
• Quantitative VOC monitoring requires reliable calibration methods.

Purpose of the Study:

• To develop a rapid machine learning (ML) calibration workflow for PID sensors.
• To map PID signals and meteorological data to VOC OH reactivity (R_OH,PTR).
• To enhance the quantitative reliability and consistency of PID sensor networks.

Main Methods:

• Co-location of four MiniPID sensors with a PTR-ToF-MS and thermohygrometer.
• Data harmonization to 10-second resolution.
• Evaluation of multiple regression models, focusing on ensemble methods (RF, XGBoost) with time-aware validation.

Main Results:

• XGBoost ensemble model demonstrated strong agreement with PTR-derived VOC OH reactivity (Pearson's r = 0.85, R^2 = 0.64).
• The ML approach significantly improved inter-sensor consistency.
• A short-duration calibration strategy was validated using out-of-time evaluation.

Conclusions:

• Rapid ML calibration effectively corrects PID sensor drift and environmental influences.
• The workflow enables practical, co-location-based harmonization of PID networks.
• This approach supports high-temporal-resolution VOC reactivity monitoring in diverse environments.