变异函数对泛克里金法的细粒子比星-地融合影响研究

泛克里金方法进行星-地融合可有效提高MODIS FMF的精度,然而由于地基站点稀少造成融合前需要利用长时间序列数据获取变异函数的主要参数（块金值、基台值和变程）,故不能满足基于卫星瞬时观测遥感PM_2.5的PMRS模型的需求。本文对2010年12月至2016年11月中国中东部地区的数据进行了变异函数参数的计算和分析,结果表明不同年份相关距离变化情况相一致,夏季显著高于冬季,基台值呈现与相关距离相反的趋势。通过利用2016年冬季变异函数中的变程（控制实验）和2011-2016年冬季变异函数的变程季均值（对比实验）作为初始值,对2016年冬季中国中东部地区MODIS FMF和地基FMF进行了融合,弃一交叉验证结果显示控制实验下FMF融合结果与地基FMF偏差最大值由0.552降低至0.198左右（对比实验下最大偏差为0.218）,平均误差相近（分别为0.070、0.080）。2种实验估算的PM_2.5平均值（分别为77.6、78.8 μg/m³）仅相差1.2 μg/m³,与在位测量的PM_2.5观测值相比,误差平均值均为37.4 μg/m³。由此可见,融合结果对初始变程值的变化敏感度不高,在季节相同的情况下,变程的多年季均值可有效替代相应季节的变程值。

Impact of Variogram Parameters on Merging Satellite and Ground-Based FMF Based on Universal Kriging

In order to improve the estimation accuracy of fine particulate matter (PM_2.5) near the surface, fine mode fraction (FMF), one of the key parameters in the PM_2.5 remote sensing method (PMRS) should be improved due to its significant error (more than 0.3). Merging MODIS FMF and ground-based FMF (AERONET&SONET) using the universal kriging (UK) method can effectively improve the accuracy of MODIS FMF over land. However, the parameters (the nugget, the sill value and the range parameter) in exponential variogram function need to be obtained using long-term MODIS FMF data because of the sparse ground-based sites, which cannot meet the need of PMRS based on instantaneous remote sensing estimates. In this study, we calculate the parameters in exponential variogram model and analyze the parameters’ variation using all MODIS data over six years from December 2010 to November 2016. Results show that the seasonal variations of correlation lengths during different years are consistent with each other. Correlation lengths in summer are significantly longer than any other three seasons while the sill values show an opposite trend, suggesting that FMF in summer has a smaller variation than the other three seasons. Furthermore, the other three seasons need more ground-based data than summer when merging MODIS FMF and ground-based FMF data. To quantify the impact of parameters in exponential variogram function on FMF fusion results and achieve instantaneous FMF fusion products, we use the range parameter in winter of 2016 (control test, denoted as CRT) and the mean value of range parameter of 6 winters over 2011-2016 (comparison test, denoted as CMP) as initial values separately to merge MODIS FMF and ground-based FMF. Leave-one-out cross-validation results show that the maximum deviation between FMF fusion results and ground-based FMF in CRT is 0.198 (in CMP is 0.218), significantly decreased compared to the maximum deviation between MODIS FMF and ground-based FMF (0.552). The mean error between FMF fusion results in CRT and ground-based FMF is close to that between FMF fusion results in CMP and ground-based FMF (0.070 vs 0.080). Then, we apply the fusion results in CRT and CMP separately to estimating PM_2.5 mass concentration near the surface in combination with the same auxiliary data such as relative humidity, the planet boundary layer height. The estimated PM_2.5 mass concentration near the surface has a slight discrepancy with a value of 1.2 μg/m³（77.6 μg/m³ vs 78.8 μg/m³）between CRT and CMP. In addition, compared with in-situ PM_2.5 mass concentration measurements, the mean error in CRT is equal to that in CMP (37.4 μg/m³ vs 37.4 μg/m³). It can be concluded that the seasonal average of range parameter for many years can be a substitute for the range parameter in the same season since the FMF fusion results and the PM_2.5 estimates are insensitive to range parameter. As a result, we can obtain instantaneous FMF fusion results to improve the estimate accuracy of PMRS when we get more satellite FMF data in the future.