An integrated low-cost air quality sensor and a multi-task calibration framework for particulate matter.

Area of Science:

• Environmental Science
• Sensor Technology
• Public Health

Background:

• Air pollution poses significant health risks, necessitating effective monitoring solutions.
• Low-cost sensors (LCSs) are promising but often lack power independence and data quality requires improvement.
• Existing calibration methods for LCSs often overlook sensor-specific and spatial factors.

Purpose of the Study:

• To develop an integrated, wireless, and self-powered LCS for simultaneous monitoring of multiple air pollutants.
• To propose a novel multi-task calibration framework using TabNet to enhance the accuracy of PM2.5 and PM10 measurements.
• To investigate the impact of sensor differences, spatial autocorrelation, and solar zenith angle on LCS calibration.

Main Methods:

• Development of an integrated wireless LCS system independent of external power.
• Implementation of a multi-task calibration framework utilizing the TabNet deep learning model.
• Inclusion of meteorological variables, sensor differences (SD), spatial autocorrelation (SA), and solar zenith angle (SZA) in the calibration models.

Main Results:

• The proposed calibration framework significantly improved R-squared values for PM2.5 (0.919 to 0.942) and PM10 (0.909 to 0.938).
• Models demonstrated robust performance through ten-fold cross-validation and successful field generalization.
• Analysis revealed that sensor proximity does not consistently correlate with improved generalization performance.

Conclusions:

• The developed wireless LCS and advanced calibration framework offer a more flexible and accurate approach to air quality monitoring.
• Incorporating sensor-specific and spatial data enhances the reliability of LCS measurements for particulate matter.
• Findings provide critical insights for optimizing urban air quality monitoring networks and related health research.