Low-Cost CO2 NDIR Sensors: Performance Evaluation and Calibration Using Machine Learning Techniques

Abstract

The study comprehensively evaluates low-cost CO₂ sensors from different price tiers, assessing their performance against a reference-grade instrument and exploring the possibility of calibration using different machine learning techniques. Three sensors (Sunrise AB by Senseair, K30 CO₂ by Senseair, and GMP 343 by Vaisala) were tested alongside a reference instrument (Los Gatos precision greenhouse gas analyzer). The results revealed differences in sensor performance, with the higher cost Vaisala sensors exhibiting superior accuracy. Despite its lower price, the Sunrise sensors still demonstrated reasonable accuracy. Meanwhile, the K30 sensor measurements displayed higher variability and noise. Machine learning models, including linear regression, gradient boosting regression, and random forest regression, were employed for sensor calibration. In general, linear regression models performed best for extrapolating data, whereas decision tree-based models were generally more useful in handling non-linear datasets. Notably, a stack ensemble model combining these techniques outperformed the individual models and significantly improved sensor accuracy by approximately 65%. Overall, this study contributes to filling the gap in intercomparing CO₂ sensors across different price categories and underscores the potential of machine learning for enhancing sensor accuracy, particularly in low-cost sensor applications.

Keywords: collocated measurements; low-cost CO2 sensors; machine learning calibration; performance evaluation.

Figure 1

Schematic representation of the sensor assembly.

Figure 2

Schematic representation of the experimental setup: (a) ambient environment test; (b) growth chamber test.

Figure 3

Boxplot showing the CO_{2_wet} and CO_{2_dry} values during the ambient environment measurements at the rooftop.

Figure 4

Time-series plot for the CO_{2_dry} measurements from all of the sensors and LGR during the ambient environment measurements from August 8, 2023 to October 14, 2023.

Figure 5

RMSE values for all three sensors and their replicates in comparison to LGR measurements before corrections and after applying corrections using different machine learning models for the testing data.

Figure 6

LGR vs. (a) Vaisala (b) Sunrise, and (c) K30 sensors showing uncalibrated CO_{2_dry} (red dots and red equation) and calibrated CO₂ (green dots and green equation) using stack ensemble machine learning for testing data.

Figure 7

RMSE values for Sunrise sensors after calibration using different machine learning models for testing data using the training dataset: (a) ambient CO₂ variations in growth chamber; (b) controlled CO₂ variations in growth chamber.