Improving Spatial Coverage of Satellite Aerosol Classification Using a Random Forest Model

Abstract

The spatial coverage of satellite aerosol classification was improved using a random forest (RF) model trained with observational data including target (aerosol type) and input (satellite measurement) variables. The AErosol RObotic NETwork (AERONET) aerosol-type dataset was used for the target variables. Satellite input variables with many missing data or low mean-decrease accuracy were excluded from the final input variable set, and good performance in aerosol-type classification was achieved. The performance of the RF-based model was evaluated on the basis of the wavelength dependence of single-scattering albedo (SSA) and fine-mode-fraction values from AERONET. Typical SSA wavelength dependence for individual aerosol types was consistent with that obtained for aerosol types by the RF-based model. The spatial coverage of the RF-based model was also compared with that of previously developed models in a global-scale case study. The study demonstrates that the RF-based model allows satellite aerosol classification with improved spatial coverage, with a performance similar to that of previously developed models.

Satellite-Based Aerosol Classification for Capital Cities in Asia Using a Random Forest Model

“…Locations of the regions of interest (Asian capital cities) are specified in Section 2.1. To classify aerosol types in Asian capital cities, we used the RF model that had been globally trained using a dataset for January 2018 to June 2020 [31], as described in Section 2.2. To test the feasibility of the RF model in Asian capital cities, we first investigated AERONET aerosol optical (spectral dependence of SSA, depolarization ratio, and FMF) and microphysical (volume size distribution and effective radius) properties for aerosol types detected by the RF model.…”

“…The minimization of missing data was found to improve spatial coverage of the RF-based aerosol classification model [31]. Compared with the previously developed model [30],~2.6 times more pixels were classified by the RF model [31] with reasonable performance (~72% accuracy). Therefore, the RF model may be a useful tool for aerosol classification even without AERONET observations.…”

“…However, misclassification of aerosols can be caused by internal uncertainties in satellite input variables. Machine learning has recently been applied in classifying aerosol types, with training of classification models with datasets comprising satellite input variables and AERONET-based aerosol types determined by Shin et al [22,30,31]. Choi et al [30] first used a random forest (RF) model to classify aerosols with an accuracy of up to 73%.…”

“…Regarding the 11 variables, four (aerosol index, solar zenith angle, CO column amount, and tropospheric NO2 column density) are from TROPOMI; and five (AOD, AE, top-of-atmosphere reflectance at three wavelengths, land cover type, and percent urban area) are from MODIS. However, the use of 11 input variables was found to have insufficient spatial coverage due to data missing for each satellite product, and an attempt was made to reduce the number of variables, thus minimizing missing data from the input variable dataset [31]. The minimization of missing data was found to improve spatial coverage of the RF-based aerosol classification model [31].…”

“…However, the use of 11 input variables was found to have insufficient spatial coverage due to data missing for each satellite product, and an attempt was made to reduce the number of variables, thus minimizing missing data from the input variable dataset [31]. The minimization of missing data was found to improve spatial coverage of the RF-based aerosol classification model [31]. Compared with the previously developed model [30],~2.6 times more pixels were classified by the RF model [31] with reasonable performance (~72% accuracy).…”