Improvement in hourly PM2.5 estimations for the Beijing-Tianjin-Hebei region by introducing an aerosol modeling product from MASINGAR
Improvement in hourly PM2.5 estimations for the Beijing-Tianjin-Hebei region by introducing an aerosol modeling product from MASINGAR
复制标题
通过引入 MASINGAR 气溶胶建模产品，改进京津冀地区每小时 PM2.5 估算
DOI：
10.1016/j.envpol.2020.114691
复制
发表时间：
2020
影响因子：
8.9
通讯作者：
Jia Li
中科院分区：
文献类型：
--
作者：
Yixiao Zhang;Wei Wang;Yingying Ma;Lixin Wu;Weiwei Xu;Jia Li
本研究通过将 Himawari-8 上先进 Himawari 成像仪的每小时气溶胶光学深度与新引入的预测器相结合，改进了传统的 PM2.5 估计模型，以估计 2018 年 11 月 1 日至 2019 年 10 月 31 日京津冀 (BTH) 地区每小时的 PM2.5 浓度。新的预测器是气溶胶种类模型的每小时 PM2.5 预测产品在全球大气中 (MASINGAR)。利用三种广泛使用的回归模型进行比较实验，即多元线性回归（MLR）、地理加权回归（GWR）和线性混合效应（LME）。十倍交叉验证 (CV) 表明 MASINGAR 产品显着提高了这些模型的性能。推出的产品提高了模型的决定系数（MLR从0.316到0.379，GWR从0.393到0.445，LME从0.718到0.765），降低了均方根误差（MLR从38.2μg/m3到36.4μg/m3，从36.0μg/m3到GWR 为 34.4 µg/m3，LME 为 24.5 µg/m3 至 22.4 µg/m3）和平均绝对误差（MLR 为 25.2 µg/m3 至 23.3 µg/m3，GWR 为 23.5 µg/m3 至 21.8 µg/m3， LME 为 15.2 µg/m3 至 13.7 µg/m3）。然后，利用训练有素的 LME 模型来估计每小时 PM2.5 浓度的空间分布。污染严重的地区集中在京津冀地区的中南部地区，污染程度最低的地区是河北西北部。每小时估算的季节性 PM2.5 水平平均值显示，冬季浓度最高 (55.4 ± 56.8 μg/m3)，夏季浓度最低 (25.1 ± 18.2 μg/m3)。 主要发现引入 MASINGAR PM2.5 产品可以将传统模型的地表 PM2.5 估算性能显着提高 7-20%。
This study improves traditional PM2.5estimation models by combining an hourly aerosol optical depth from the Advanced Himawari Imager onboard Himawari-8 with a newly introduced predictor to estimate hourly PM2.5concentrations in the Beijing–Tianjin–Hebei (BTH) region from November 1, 2018 to October 31, 2019. The new predictor is an hourly PM2.5forecasting product from the Model of Aerosol Species IN the Global AtmospheRe (MASINGAR). Comparative experiments were conducted by utilizing three extensively used regression models, namely, multiple linear regression (MLR), geographically weighted regression (GWR), and linear mixed effects (LME). A ten-fold cross validation (CV) demonstrated that the MASINGAR product significantly improved the performances of these models. The introduced product increased the model’s determination coefficients (from 0.316 to 0.379 for MLR, from 0.393 to 0.445 for GWR, and from 0.718 to 0.765 for LME), decreased their root mean square errors (from 38.2 μg/m3to 36.4 μg/m3for MLR, from 36.0 μg/m3to 34.4 μg/m3for GWR, and from 24.5 μg/m3to 22.4 μg/m3for LME) and mean absolute errors (from 25.2 μg/m3to 23.3 μg/m3for MLR, from 23.5 μg/m3to 21.8 μg/m3for GWR, and from 15.2 μg/m3to 13.7 μg/m3for LME). Then, a well-trained LME model was utilized to estimate the spatial distributions of hourly PM2.5concentrations. Highly polluted localities were clustered in the central and southern areas of the BTH region, and the least polluted area was in northwestern Hebei. Seasonal PM2.5levels averaged from the hourly estimations exhibited the highest concentrations (55.4 ± 56.8 μg/m3) in the winter and lowest concentrations (25.1 ± 18.2 μg/m3) in the summer.Main findingIntroducing the PM2.5products from MASINGAR can significantly improve the performance of traditional models for surface PM2.5estimations by 7–20%.