Development and evaluation of a robust temperature sensitive algorithm for long term NO2 gas sensor network data correction

Low-cost sensors (LCS) for air-quality monitoring have shown huge potential in enhancing spatial and temporal resolutions at a lower cost, greater flexibility in use and with less maintenance than fixed-air quality monitoring stations (AQMS). There are numerous reports showing sensor-based systems perform well under laboratory conditions. However, the accuracy of LCS operated under field conditions has been reported to deviate from their use in these controlled environments. Previous studies have employed various mathematical techniques to improve system performance. These include classic multivariate regression model (MLR) to incorporate ambient factors (e.g. temperature & relative humidity) and machine learning (ML) models for integrating complex nonlinear sensor responses for gaseous pollutant concentration predictions. However, routine field calibrations may cause data inconsistency during measurement periods. This study first illustrates the limitations of short-term routine calibration by comparing performances of existing MLR and ML models. Further, based on a long-term evaluation period, a newly principle-based method named Temperature Look-Up (TLU) model was built and compared with the existing MLR and ML models. Measurements were taken for 8 months (August 2017 to March 2018) in SanMenXia (SMX) city, China, using 8 sets of sensor systems (Mini Air Stations, MAS). One MAS was co-located with AQMS to provide data to assess the principle, whereas the other 7 MAS were used for verification. The TLU model showed the best performance for long term validation (4 months) and high model inner coherence (R-2 > 0.91). Meanwhile, the TLU model also shows the fast convergence (40 days) to reach stable prediction performance, and spatial representativeness in the sensor network. Recommendations on sensor network deployment and strategy for data maintenance based on our observations are discussed.