随着数据采集技术的进步，具有时间和空间属性的数据规模增长极快，同时其自身的时空相依结构也越来越复杂，现有分析方法受到很大挑战。本项目将从方法学上探索海量时空数据的统计建模及相关问题，并应用于我国大气污染的数据分析当中。利用空间统计学、函数型数据分析、随机过程、数据降维等领域中较新的理论和技术，本项目针对“时空结构的统计推断”、“分层模型的构建与参数估计”和“动态模型的拓展与数据融合”等几个时空分析当中的基本问题，分别从时空随机场的可分性检验、静态与动态时空模型的构造与数据融合、非线性动态模型的误差分析等方面开展研究。此外，本项目还将分析我国2013年以来多个大气污染物的海量数据，并以绘制动态污染地图和改进重污染预报精确度为切入点，从统计学的视角来理解大气污染物的扩散机制和空间传输特点。

With the growing capability of data collection, there is a rapid increase of the scale of data with complex spatial and temporal dependence structures.Statistical anaysis is thus very challenging. By using recent theories and techniques in spatial statistics, functional data analysis, stochastic process and dimension reduction, this project aims to develop new statistical modelling frameworks for massive spatiotemporal data. We will focus on several fundamental issues in spatiotemporal analysis, such as ‘statistical inferences for space-time structures’, ‘hierarchical models and parameter estimations’ and ‘extensions of dynamic models and data fusion’. In particular, we will mainly investigate testing methods of the separability of space-time random fields, develop new static models and data fusion procedures, and extend them to dynamic versions so that the error in nonlinear dynamic systems can be better controlled. Moreover, we will apply our new models on multiple air pollutants data that have been collected all over China since year 2013. Through making dynamic pollution maps and improving forecasting powers of heavy pollutions, the application will provide us more statistical insights in the diffusion and transport mechanisms of air pollutants.