High-performance machine-learning-based calibration of low-cost nitrogen dioxide sensor using environmental parameter...

Area of Science:

• Environmental Science
• Sensor Technology
• Data Science

Background:

• Accurate monitoring of harmful gases like nitrogen dioxide (NO2) is crucial for mitigating air pollution's environmental and health impacts.
• Existing NO2 monitoring equipment is often expensive and complex, necessitating more affordable and reliable alternatives.
• Urban NO2 pollution, primarily from fossil fuel combustion, poses significant risks to respiratory health.

Purpose of the Study:

• To develop and validate a novel, cost-effective method for calibrating low-cost NO2 sensors.
• To integrate machine learning with advanced data processing techniques for enhanced sensor accuracy.
• To provide a dependable alternative for widespread NO2 monitoring in urban environments.

Main Methods:

• Implemented a machine learning approach combining neural network surrogates and global data scaling.
• Utilized expanded correction model inputs, including environmental parameter differentials and data from multiple NO2 sensors.
• Validated the methodology using a purpose-built platform and comparative experiments against high-precision reference stations over five months.

Main Results:

• The calibrated low-cost NO2 sensors achieved remarkable correction quality, with a correlation coefficient exceeding 0.9 against reference data.
• The root mean squared error was below 3.2 µg/m³, demonstrating high accuracy.
• The developed method proved effective across various calibration scenarios and input configurations.

Conclusions:

• The proposed machine learning-based calibration method offers a dependable and cost-effective solution for NO2 monitoring.
• This approach significantly enhances the reliability of low-cost sensors, making them a viable alternative to expensive stationary equipment.
• The findings support the broader implementation of accessible air quality monitoring networks.