An Efficient and Accurate Model Coupled With Spatiotemporal Kalman Filter and Linear Mixed Effect for Hourly PM2.5 Mapping

The ground PM2.5 mapping method based on aerosol optical depth (AOD) is universal. However, since the modeling samples for hourly PM2.5 mapping are usually very large, and it also lacks an efficient spatiotemporal model to describe the strong spatiotemporal correlation that existed in the large hourly PM2.5 modeling samples, most of the existing studies focus on the daily, monthly, and annual average scales. In this study, a model coupled with a spatiotemporal Kalman filter and linear mixed effect (LMESTKF) is proposed. Based on the hourly seamless 1-km AOD data generated by fusing multiple satellite products and numerical model data products, we realize efficient and accurate hourly seamless 1-km PM2.5 mapping in the Beijing-Tianjin-Hebei (BTH) urban agglomeration region. Results show that the goodness of fit (
$R^{2}$
), root mean square error (RMSE), bias, and mean absolute percentage error (MAPE) metrics based on sample- and site-based tenfold cross-validation method are (0.91,
$14.37 \mu \text{g}/\text{m}^{3}$
,
$- 0.41 \mu \text{g}/\text{m}^{3}$
, 28.91%) and (0.87,
$16.98 \mu \text{g}/\text{m}^{3}$
,
$- 0.32 \mu \text{g}/\text{m}^{3}$
, 34.62%), respectively; the accuracy is higher than those of the existed PM2.5 mapping models in the BTH urban agglomeration region. In addition, the time and memory consumption required to construct the LMESTKF model for hourly PM2.5 mapping at the monthly scale is only around 10 min and 4 GB, respectively, indicating that the LMESTKF model proposed in this study is not only high in accuracy but also fast in efficiency. These results prove that the proposed LMESTKF model can efficiently generate accurate hourly PM2.5 maps, which can further promote the prevention and control of PM2.5 pollution in China.