Mapping nighttime PM2.5 from VIIRS DNB using a linear mixed-effect model

Estimation of particulate matter with aerodynamic diameter less than 2.5 mu m (PM2.5) from daytime satellite aerosol products is widely reported in the literature; however, remote sensing of nighttime surface PM2.5 from space is very limited. PM2.5 shows a distinct diurnal cycle and PM2.5 concentration at 1:00 local standard time (LST) has a linear correlation coefficient (R) of 0.80 with daily-mean PM2.5. Therefore, estimation of nighttime PM2.5 is required toward an improved understanding of temporal variation of PM2.5 and its effects on air quality. Using data from the Day/Night Band (DNB) of the Visible Infrared Imaging Radiometer Suite (VIIRS) and hourly PM2.5 data at 35 stations in Beijing, a mixed-effect model is developed here to estimate nighttime PM2.5 from nighttime light radiance measurements based on the assumption that the DNB-PM2.5 relationship is constant spatially but varies temporally. Cross-validation showed that the model developed using all stations predict daily PM2.5 with mean determination coefficient (R-2) of 0.87 +/- 0.12, 0.83 +/- 0.10, 0.87 +/- 0.09, 0.83 +/- 0.10 in spring, summer, autumn and winter. Further analysis showed that the best model performance was achieved in urban stations with average cross-validation R-2 of 0.92. In rural stations, DNB light signal is weak and was likely smeared by lunar illuminance that resulted in relatively poor estimation of PM2.5. The fixed and random parameters of the mixed-effect model in urban stations differed from those in suburban stations, which indicated that the assumption of the mixed-effect model should be carefully evaluated when used at a regional scale.

文献中广泛报道了利用日间卫星气溶胶产品估算空气动力学直径小于2.5微米的颗粒物（PM2.5），然而，从太空对夜间地表PM2.5的遥感非常有限。PM2.5呈现出明显的昼夜循环，当地标准时间1点的PM2.5浓度与日平均PM2.5的线性相关系数（R）为0.80。因此，为了更好地理解PM2.5的时间变化及其对空气质量的影响，需要对夜间PM2.5进行估算。利用可见光红外成像辐射仪组（VIIRS）的日夜波段（DNB）数据以及北京35个站点的每小时PM2.5数据，基于DNB - PM2.5关系在空间上恒定但在时间上变化的假设，在此建立了一个混合效应模型，通过夜间光辐射测量来估算夜间PM2.5。交叉验证表明，使用所有站点开发的模型预测每日PM2.5，在春季、夏季、秋季和冬季的平均决定系数（R²）分别为0.87±0.12、0.83±0.10、0.87±0.09、0.83±0.10。进一步分析表明，在城市站点模型性能最佳，平均交叉验证R²为0.92。在农村站点，DNB光信号较弱，可能受到月球光照的干扰，导致PM2.5估算相对较差。城市站点混合效应模型的固定参数和随机参数与郊区站点不同，这表明在区域尺度使用混合效应模型时，应仔细评估其假设。