Performance evaluation of low-cost atmospheric particulate matter sensors and comparison of calibration methods

Low-cost particulate matter sensors can supplement limited data from air quality monitoring stations. Still, the accuracy of the measurement results is easily affected by environmental factors and often needs to be corrected before use. Previous studies have focused on single PM_2.5 data and ignored the exploration of PM₁₀ data. Therefore, this paper evaluates the accuracy of PM_2.5 and PM₁₀ and compares common correction methods. First, the sensor’s performance is assessed by comparing it with the data reported by regulatory-level instruments. Then, the impact of different factors on measurement errors (relative humidity, temperature, particle number, and human activity time) is analyzed. Finally, six correction models are compared separately. Results are as follows: PM_2.5 measurement accuracy is high: The correlation coefficient R² is 0.8, and the Root mean square error (RMSE) is 10.6 μg/m%. Even without correction, it has good applicability. Simple linear model correction can achieve high credibility, and complex statistical models perform better than traditional methods. PM₁₀ measurement data has low accuracy R²=0.32, RMSE=31.7 μg/m³. As the relative humidity increases, the absolute value of the error decreases, and as the temperature increases, the absolute value of the error increases. Without correction, it cannot be used as a reference. Simple linear correction can improve the PM₁₀ data quality to a certain extent, but only after correction by a complex statistical model can high-reliability data be provided. The random forest model has the best correction effect. These results provide a reference for low-cost sensors and have practical guiding significance.