基于历史信息交互时空模型的长时间PM2.5重建方法研究

利用卫星遥感反演的AOD（气溶胶光学厚度）估算地面PM2.5（细颗粒物）浓度是获取PM2.5污染信息的重要方法。但由于缺乏PM2.5-AOD模型开发校对所必须的地面观测值，针对历史时期(2013年以前)的相关研究较少。如何改善历史时期PM2.5-AOD日均关系成为基于卫星遥感重建长时期地面PM2.5浓度的重点和难点，也是提高基于卫星遥感的大气污染估算精度的重要科学问题。本研究拟从改善高分辨率AOD的缺失问题出发，在探索影响因素间相互作用来选取建模因子增强PM2.5-AOD关系的基础上，通过在时空权重机制中引入影响因素在历史时期和建模时期的相似性来改造地理时空加权回归模型（GTWR），重建历史长时期PM2.5数据，旨在提供一种适用于缺乏依赖变量观测值约束时期的统计建模方法，提高PM2.5的估算精度。本项目的开展将有助于克服PM2.5污染信息获取的“瓶颈”，具备切实的科学研究意义和实用价值。

Ambient concentrations of fine particulate matter (PM2.5) in a long-term period are substantially required for exploring dynamical pattern of PM2.5 pollution and assessing the health effect of PM2.5 exposure. However, few studies have conducted satellite-based PM2.5 estimation for China before 2013, due to the limited PM2.5 measurements available from ground-level monitoring stations. To extend the temporal coverage of PM2.5 observations, this project proposes a spatiotemporal modeling method to quantify the relationship between surface PM2.5 and satellite aerosol optical depth (AOD) and predict daily high-resolution PM2.5 concentration in a long-term period of 2003 to 2012. First, to improve the missingness of AOD, a fused 1-km AOD dataset would be generated by merging the newly Multi-Angle Implementation of Atmospheric Correction (MAIAC) AOD data and other coarse-resolution aerosol products. The use of high-resolution AOD data would represent more aerosol gradients and directly benefit the relationship of AOD and PM2.5. Then, “higher-order” impact factors, identified through analyzing the physical PM2.5-AOD relationship, would be used as explanatory variables in modeling. Unlike previous PM2.5-AOD studies only applies the “first-order” factors, high-order factor such as AOD2 and AOD×RH (Relative Humidity) can help interpret the interaction between PM2.5 and its influence factors. Then, a space-time regression model that is a geographically and temporally weighted regression (GTWR) model improved with weighting mechanism using dynamical historical information would be devised to predict historical PM2.5 concentrations for each day. Through considering the difference in daily PM2.5-AOD relationship between historical and modeling years, it is possible for the improved GTWR model to generate better daily PM2.5 predictions before 2013. Thus, this project proposes a promising spatiotemporal PM2.5-AOD model to improve daily accuracy of historical PM2.5 estimates and extends the temporal coverage of PM2.5 observations.