TY  - JOUR
AU  - Jiang, Mengjiao
AU  - Chen, Zhihang
AU  - Yang, Yinshan
AU  - Ni, Changjian
AU  - Yang, Qi
PY  - 2022
DA  - 2022//
TI  - Establishment of aerosol optical depth dataset in the Sichuan Basin by the random forest approach
JO  - Atmospheric Pollution Research
SP  - 101394
VL  - 13
IS  - 5
KW  - Aerosol optical depth, Dataset, Random forest, Cloudy areas
AB  - The Sichuan Basin has become one of the four city clusters and heavy polluted regions in China. In this study, the random forest (RF) machine learning method and multiple datasets are used to establish aerosol optical depth (AOD) dataset in the cloudy Sichuan Basin. Multiple datasets include ground-based PM10 and PM2.5, the AOD from the Sun-sky radiometer Observation Network (SONET) and the Second Modern-Era Retrospective analysis for Research and Applications (MERRA-2) aerosol reanalysis, and several meteorological variables. The correlation analysis, variance inflation factor method, covariance test, and important scores are used to select variables for the model. Eight independent variables, including MERRA-2 AOD, PM10, PM2.5/PM10, low cloud cover, 2 m air temperature, relative humidity, wind direction and boundary layer height, and one dependent variable SONET AOD are selected for the model in Chengdu, the capital of Sichuan, and then extended to the Sichuan Basin. The 10-fold cross validation and statistical comparison of the Multi-Angle implementation of Atmospheric Correction (MAIAC) and the MERRA-2 AOD are conducted. Results show that the values of PM10 and PM2.5, and MERRA-2 AOD are highest at the bottom of the basin, followed by that at the edge of the basin, and the lowest at the plateau areas. Comparing with the SONET AOD, the MERRA-2 and MAIAC underestimate the AOD in the Sichuan Basin, with the linear regression slope of 0.57 and 0.74, respectively. The RF AOD shows the best accuracy with the 10-fold cross-validation correlation coefficient of 0.79, the smallest RMSE of 0.17 and MAE of 0.14.
SN  - 1309-1042
UR  - https://www.sciencedirect.com/science/article/pii/S1309104222000800
UR  - https://doi.org/https://doi.org/10.1016/j.apr.2022.101394
DO  - https://doi.org/10.1016/j.apr.2022.101394
ID  - JIANG2022101394
ER  -