Using Multi-parameters for Calibration of Low-cost Sensors in Urban Environment

Measurements that were taken at appropriate spatial and temporal resolution are important for understanding urban environment. However, due to cost issues, most of current monitoring sensors are sparsely deployed and are not able to provide sufficient spatial resolution. As an alternative solution, low-cost sensors that cost several orders less than the current sensors have been exploited, providing much higher spatial resolution with relatively low cost. However, the data from low-cost sensors are widely reported to be deficient, resulting in the calibration of low-cost sensors being more difficult. In this work, key challenges of calibration of low-cost sensors were identified, and the limitations of current calibration methods were discussed. A multi-parameter calibration that not only utilises cross-sensitive parameters but also considers other relevant parameters was proposed. The stepwise regression method with interaction term was then proposed to systematically select optimal parameters for the calibration. The evaluations that were carried out in both city centre and outside of city centre have shown a great advantage of using the proposed method. It shows a significant better result than the existing methods, in terms of improved root mean square errors and better linearity between the calibrated trace and the reference.