This study develops a set of research processes to extract the spatiotemporal mechanism of PM2.5 diffusion. In the set of research processes, the first part is the innovative data fusion technology to improve the spatiotemporal resolution of PM2.5 estimation, and the second part is the PM2.5 diffusion feature extraction technology. The data of the air pollution season of central Taiwan from 2010 to 2018 was adopted as a demonstration, the PM2.5 diffusion characteristics and the corresponding atmospheric conditions on the pollution days (PM2.5 daily average >35μg/m3) were discussed.The research integrated the IoT and spatiotemporal big data analysis technology to explore the PM2.5 spatiotemporal feature and physical mechanism. Without losing data quality, the spatiotemporal resolution of PM2.5 estimation was improved through innovative data fusion technology. Additionally, we pioneered to adopt signal analysis methods combined with machine learning methods to extract the main patterns of PM2.5 diffusion distribution. The atmospheric diffusion conditions, estimated by several weather factors, assisted to link the corresponding diffusion patterns under different weather conditions. The set of research processes is able to apply in other regions of Taiwan, and it is expected to establish Taiwan air pollution warning indicators based on atmospheric diffusion conditions.The research integrates IoT, big data analysis, and GIS, and can be applied to the whole Taiwan or even the world. The set of research processes allows companies, government agencies, schools, and research institutes to present the users of visualized pollutant analysis results through various platforms. Additionally, the more effective scientific reference basis is provided to formulating air quality control policies, urban planning, assessment of health risk exposure, arrangement and development of microsensors, etc.