The improvement of spatial-temporal resolution of PM_2.5 estimation based on micro-air quality sensors by using data fusion technique

With the rapid development of the Internet of things (IoTs) and modern industrial society, forecasting air pollution concentration, e.g., the concentration of PM_2.5, is of great significance to protect human health and the environment. Accurate prediction of PM_2.5 concentrations is limited by the number and the data quality of air quality monitoring stations. In Taiwan, the spatial and temporal data of PM_2.5 concentrations are measured by 77 national air quality monitoring stations (built by Taiwan EPA). However, the national stations are costly and scarce because of the highly precise instrument and their size. Therefore, many places are still out of coverage of the monitoring network. Recently, under the framework of IoTs, there are hundreds of portable air quality sensors called “AirBox” developed jointly by the Taiwan local government and a private company. By virtue of its low price and portability, the AirBox can provide a higher resolution of space-time PM_2.5 measurement. However, the spatiotemporal distribution is different between AirBox and EPA stations, and data quality and accuracy of AirBox is poorer than national air quality monitoring stations. Thus, to integrate the heterogeneous PM_2.5 data, the data fusion technique should be used before further analysis. In this study, we propose a new data fusion method called multi-sensor space-time data fusion framework. It is based on the Optimum Linear Data Fusion theory and integrating with a multi-time step Kriging method for spatial-temporal estimation. The method is used to do heterogeneous data fusion from different sources and data qualities. It is able to improve the estimation of PM_2.5 concentration in space and time. Results have shown that by combining PM_2.5 concentration data from 1176 low-cost AirBoxes as additional information in our model, the estimation of spatial-temporal PM_2.5 concentration becomes better and more reasonable. The r² of the validation regression model is 0.89. Under the approach proposed in this study, we made the information of the micro-sensors more reliable and improved the higher spatial-temporal resolution of air quality monitoring. It could provide very useful information for better spatial-temporal data analysis and further environmental management, such as air pollution source localization, health risk assessment, and micro-scale air pollution analysis.