基于监测及Kriging方法的京津冀地区大气污染物暴露分布研究

为了实现充分利用已有环境监测站点数据进行人群精细化暴露评估的目的，同时解决某些待测人群社区周边无监测站点时数据的选择问题，以保定市作为大气高污染研究城市，基于现场监测和Kriging（克里金插值）空间分析方法，明确了在研究大气污染物人群暴露时，某一个固定监测站污染物数据的代表性问题.研究表明:对于大气中φ（SO₂）、φ（NO₂）、颗粒物及其组分，空气质量监测点位的代表性一般为5~6 km；对于φ（CO）、φ（O₃）和φ（VOCs），它们在城市不同地区的空间分布更为均匀，空气质量监测点位的代表性范围更大.通过使用Radial Basis Functions（径向基函数，RBF）、Local Polynomial Interpolation（局部多项式插值，LPI）、Inverse Distance Weighting（反距离权重插值，IDW）、Kriging、Kernel Smoothing（内核平滑插值，KS）和Diffusion Kernel（内核扩散插值，DK）等6种空间分析方法对大气污染物浓度进行预测发现，Kriging方法对大气污染物浓度预测时可使预测值和实测值间的偏差小于10%，准确度最高.因此，在进行某城市某点位的污染物人群暴露浓度预测时，若该点位周边5 km以内有空气质量监测点位，则可用该点位的监测值代替；若5 km以内没有空气质量监测点位，则可基于最近监测点位的污染物浓度进行Kriging空间插值，从而获得该点位的污染物暴露水平. 

Exposure Distribution of Air Pollutants in Beijing-Tianjin-Hebei Region Based on Monitoring and Kriging Method

In order to make full use of the existing environmental monitoring site data for refined population exposure assessment, and to solve the problem of choosing data when there are no monitoring sites around certain communities of the population to be measured, Baoding was taken as a research city with heavy air pollution. Based on the field monitoring in Baoding and the simulation by the Kriging method, the representativeness of a stationary monitoring site in the study of air pollutant exposure levels was determined. The results showed that for φ(SO₂), φ(NO₂), particulate matter and its main composition, the data representativeness of a stationary monitoring site was generally about 5-6 km, while for φ(CO), φ(O₃) and φ(VOCs), the spatial distribution in different areas was more uniform and the representativeness was larger. A comparation of six methods, including Radial Basis Functions, Local Polynomial Interpolation, Inverse Distance Weighting, Kriging, Kernel Smoothing and Diffusion Kernel, showed that the Kriging method had the highest predictive accuracy (the prediction error less than 10%). In conclusion, if there is a stationary monitoring site within 5 km, the monitoring concentrations can be used as the exposure levels to the air pollutants. The air pollutant levels beyond 5 km can be simulated using Kriging spatial interpolation based on the data at the nearest available station.