# Effects of Environmental Factors on the Performance of Ground-Based Low-Cost CO2 Sensors

## TL;DR

This study improves low-cost CO2 sensors by using a calibration method that accounts for environmental factors like temperature and humidity.

## Contribution

A multivariable linear regression calibration method is introduced to significantly reduce CO2 sensor bias and error.

## Key findings

- In lab conditions, calibration reduced mean absolute bias from 5.218 ppm to 0.003 ppm.
- Field observations showed a decrease in RMSE from 8.315 ppm to 2.154 ppm after calibration.
- Calibration windows from winter or summer seasons yield better results for sensor accuracy.

## Abstract

This paper presents a multivariable linear regression calibration method for non-dispersive infrared (NDIR) CO2 sensors in a low-cost carbon monitoring network. We test this calibration method with data collected in a temperature- and pressure-controlled laboratory and evaluate the calibration method with long-term observational data collected at the Xinglong Atmospheric Background Observatory. Compared to data collected by a high-accuracy cavity ring-down spectrometer (Picarro), the results show that a multivariable linear regression approach incorporating temperature, pressure, and relative humidity can reduce the mean absolute bias from 5.218 ppm to 0.003 ppm, with root mean square errors (RMSE) within 2.1 ppm after calibration. For field observations, the RMSE is reduced from 8.315 ppm to 2.154 ppm, and the bias decreases from 39.170 ppm to 0.018 ppm. The calibrated data can effectively capture the diurnal variation of CO2 mole fraction. The test of the number of reference data shows that about 10 days of co-located reference data are sufficient to obtain reliable measurements. Calibration windows taken from winter or summer provide better results, suggesting a strategy to optimize short-term calibration campaigns.