多尺度PM_(2.5)分布特征的空间插值与遥感反演对比

Contrastive Analysis of Multi-scale PM_(2.5) Concentration Spatial Distribution Simulation of Spatial Interpolation and Remote Sensing Inversion

精确识别污染物浓度的空间分布是进行区域大气污染防治的重要基础。利用MODIS卫星数据,采用基于地面气象和环境空气质量监测站点观测数据为基础的反演模型,反演获取2013年12月珠三角地区典型大气污染过程1 km分辨率的PM_(2.5)浓度数据,对比分析遥感反演及基于环境空气质量监测站点观测数据的空间插值方法对区域、城市和乡镇尺度PM_(2.5)浓度空间分布特征的再现效果差异。结果表明,珠三角地区PM_(2.5)遥感反演结果与地面观测数据的相关性达到0.74,相关性水平较好,遥感反演结果可描述区域、城市和乡镇尺度上PM_(2.5)污染浓度的空间分布特征,识别不同空间位置的污染程度差异;基于站点观测数据的空间插值方法对PM_(2.5)浓度空间分布特征的再现能力有限,在区域尺度PM_(2.5)浓度空间分布特征分析时效果尚可,在站点有限的城市和乡镇尺度分析中效果不佳,容易产生对高浓度污染地区的误判;在需要利用站点观测数据分析区域尺度PM_(2.5)浓度空间分布特征时,析取克里金、反距离权重或径向基函数插值方法的效果相对较好。

Identification of the pollutant concentration spatial distribution is an important foundation of regional atmospheric pollution control. 1 km resolution PM（2.5） concentration maps of Pearl River Delta region（PRD） are prepared through MODIS satellite data inversion. Taking a typical pollution day in December 2013 as a study case, multi-scale contrastive analysis between remote sensing inversion method and different spatial interpolation methods based on site observation data are carried out to compare the ability and applicability in representing the spatial distribution characteristics of atmospheric PM（2.5） concentration. Results show that remote sensing inversion method could effectively characterize the spatial distribution of atmospheric PM（2.5） concentration in township, city and regional scales, identifying spatial pollution variations. The spatial distribution characteristic representing ability and applicability of spatial interpolation methods based on site observation data are limited. Some of the spatial interpolation methods can be effective in regional scale analysis. But no spatial interpolation methods are effective in city and township scale analysis, or could effectively estimate the pollutant concentration level where the monitoring sites are sparse. When the spatial interpolation methods have to be used to estimate the spatial distribution pattern of regional atmospheric PM（2.5） concentration, disjunctive kriging, inverse distance weighting or radial basis function method could be better choices.