ABSTRACT The physical processes associated with the constituents of the troposphere, such as aerosols have an immediate impact on human health. This study employs a novel method to calibrate Aerosol Optical Depth (AOD) obtained from the MODerate resolution Imaging Spectrometer (MODIS – Terra satellite) for estimating surface PM2.5 concentration. The Combined Deep Blue Deep Target daily product from the MODIS AOD data acquired across the Indian Subcontinent was used as input, and the daily averaged PM2.5pollution level data obtained from 33 monitoring stations spread across the country was used for calibration. Mixed Effect Models (MEM) is a linear model to deal with non-independent data from multiple levels or hierarchy using fixed and random effects of dependent parameters. MEM was applied to the dataset obtained for the period from January to August 2017. The MEM considers a fixed and random component, where the random components model the daily variations of the AOD – PM2.5 relationships, site-specific adjustment parameters, temporal (meteorological) variables such as temperature, and spatial variables such as the percentage of agricultural area, forest cover, barren land and road density with the resolution of 10 km × 10 km. Estimation accuracy was improved from an R2 value of 0.66 from our earlier study (when PM2.5 was modeled against only AOD and site-specific parameters) toR2 value of 0.75 upon the inclusion of spatiotemporal (meteorological) variables with increased % within Expected Error from 18% to 35%, reduced Mean Bias Error from 3.22 to 0.11 and reduced RMSE from 29.11 to 20.09. We also found that spline interpolation performed better than IDW and Kriging inefficiently estimating the PM2.5 concentrations wherever there were missing AOD data. The estimated minimum PM2.5 is 93 ± 25μg/m3 which itself is in the upper limit of the hazardous level while the maximum is estimated as 170 ± 70μg/m3. The study has thus made it possible to determine the daily spatial variations of PM2.5 concentrations across the Indian subcontinent utilizing satellite-based AOD data.

中文翻译：

---

用于估计印度次大陆 MODIS AOD PM2.5 的时空混合效应模型

摘要 与对流层成分相关的物理过程，如气溶胶，对人类健康有直接影响。本研究采用了一种新方法来校准从中等分辨率成像光谱仪（MODIS – Terra 卫星）获得的气溶胶光学深度 (AOD)，以估计表面 PM2.5 浓度。以整个印度次大陆获取的 MODIS AOD 数据的组合 Deep Blue Deep Target 日积作为输入，并使用从遍布全国的 33 个监测站获得的日平均 PM2.5 污染水平数据进行校准。混合效应模型 (MEM) 是一种线性模型，用于使用相关参数的固定和随机效应处理来自多个级别或层次结构的非独立数据。MEM 应用于 2017 年 1 月至 8 月期间获得的数据集。 MEM 考虑固定和随机分量，其中随机分量模拟 AOD - PM2.5 关系的每日变化、特定地点的调整参数、时间（气象）温度等变量，以及农业面积百分比、森林覆盖率、荒地和道路密度等空间变量，分辨率为 10 km × 10 km。估计精度从我们早期研究中的 R2 值 0.66（当 PM2.5 仅针对 AOD 和特定地点参数建模时）提高到 0.75，在包含时空（气象）变量后，在预期误差范围内增加了 % 18% 到 35%，平均偏差误差从 3.22 减少到 0.11，RMSE 从 29.11 减少到 20.09。我们还发现，在缺少 AOD 数据的情况下，样条插值比 IDW 和克里金法在估算 PM2.5 浓度方面效率低下。PM2.5 的估计最小值为 93±25μg/m3，其本身处于危险水平的上限，而最大值估计为 170±70μg/m3。因此，该研究使得利用基于卫星的 AOD 数据确定印度次大陆 PM2.5 浓度的每日空间变化成为可能。