基于时空大数据的中国区域小时级精细分辨率PM2.5浓度分布研究

The heavy fine particulate matters (PM2.5) pollution in China not only affects life style, but also threats people’s health. It is very important for the health effect evaluation and public policy making to obtain accurate spatial and temporal distribution of PM2.5. Near ground monitoring stations of air quality can only represent a small portion of areas. They are not sufficient for large scale and detail research. Aerosol optical depth (AOD) from remote sensing provide a good way to calculate the near ground PM2.5 mass density. However, there are often a large number of missing pixels in the AOD product due to the effect of cloud, ground with high reflection, and so on. In this project, we propose an efficient model to generate high covered hourly PM2.5 mass concentration with high spatial resolution in China. Considering the heterogeneous distribution of PM2.5 due to the different pollutant source, complicated land surface and climate environment, we will propose a Spatial-Temporal Adaptive Regression model with Non-homogeneity (STARN) by integrating both spatial autocorrelation and non-homogeneity. It will involves two kinds of non-homogeneity, i.e. parameter difference of local models and nonlinear effect between dependent variables and independent variables. Based on the STARN model, we then research a deformation method for the spatial-temporal non-stationary random field to obtain a new spatial-temporal random field with stationarity in the mean and second-order momentum. A strategy of spatial-temporal trend decomposition will be proposed to decompose the PM2.5 mass concentration into spatial-temporal trend and residue. A spatial-temporal variogram model which consider some critical time frame will then be constructed from the residue part to model the spatial-temporal autocorrelation. PM2.5 mass concentration distribution can be estimated by spatial-temporal regression Kriging model and downscaled by area-to-point Kriging.

PM2.5污染不仅严重影响人们的生活方式，而且直接威胁到人体健康，准确掌握其时空分布是精准估算个体污染暴露量、评估健康影响和制定合理政策的关键因素。数量有限的地面监测点无法满足精细尺度应用需求，基于遥感的AOD产品为大范围PM2.5浓度反演提供了可能，但其时空分辨率低。针对我国各地污染源、气象环境和地表类型等差异带来的PM2.5非均质分布情况，本研究基于非平稳时空随机场模型提出一套适合于我国复杂下垫面的小时级1Km分辨率PM2.5浓度分布生成方法。提出耦合时空相关性和异质性的时空自适应非线性回归模型，重点研究模型局部参数差异和影响因素非线性关系两种异质性；提出以关键“时间帧”为重要参照的时空自相关性度量方法，突破非平稳时空随机场插值方法，在形变空间内建立时空相关性变异函数，建立时空回归克里格模型，实现污染物浓度分布最优无偏估计。最后，通过面到点降尺度模型实现空间分辨率提升。

PM2.5污染不仅严重影响人们的生活方式，而且直接威胁到人体健康，准确掌握其时空分布是精准估算个体污染暴露量、评估健康影响和制定合理政策的关键因素。数量有限的地面监测点无法满足精细尺度应用需求，基于遥感的AOD产品为大范围PM2.5浓度反演提供了可能，但其时空分辨率低。针对我国各地污染源、气象环境和地表类型等差异带来的PM2.5非均质分布情况，本研究基于非平稳时空随机场模型提出一套适合于我国复杂下垫面的小时级1Km分辨率PM2.5浓度分布生成方法。提出耦合时空相关性和异质性的时空自适应非线性回归模型，重点研究模型局部参数差异和影响因素非线性关系两种异质性；提出以关键“时间帧”为重要参照的时空自相关性度量方法，突破非平稳时空随机场插值方法，在形变空间内建立时空相关性变异函数，建立时空回归克里格模型，实现污染物浓度分布最优无偏估计。最后，通过面到点降尺度模型实现空间分辨率提升。