Publications

Jia, SQ; Han, M; Zhang, CK (2022). Long short-term memory network model to estimate PM2.5 concentrations with missing-filled satellite data in Beijing. STOCHASTIC ENVIRONMENTAL RESEARCH AND RISK ASSESSMENT, 36(12), 4175-4184.

Abstract
Fine particulate matter (PM2.5) concentrations pollution is one of serious environmental issues. It is necessary for PM2.5 concentrations estimation because the existing PM2.5 ground monitoring stations are relatively sparse and cannot obtain continuous PM2.5 concentrations over a large area. Several studies have been applied aerosol optical depth (AOD) to PM2.5 concentrations estimation. However, the missing of the AOD data does not improve the accuracy of PM2.5 estimations. Therefore, we need a filling technique to fill the AOD data. The purpose of this study is to deal with the missing AOD data using machine learning interpolation techniques and to estimate the PM2.5 concentrations at 1 km resolution by adding auxiliary factors to fit the relationship between AOD data and PM2.5 data. We used a long short-term memory network (LSTM), model to fit the filled AOD data. We estimated PM2.5 values based on the meteorological conditions and the AOD data at the station, and we also established the temporal and spatial relationships of PM2.5. Overall, the method is suitable for PM2.5 estimations with R-2 = 0.75. We conducted experiments at existing stations in Beijing. The results of the study demonstrate the validity of gap-filling AOD data for PM2.5 estimations, with PM2.5 distributions being higher in the south and lower in the north.

DOI:
10.1007/s00477-022-02253-8

ISSN:
1436-3259