TY - JOUR
T1 - Using multi-angle imaging spectro radiometer aerosol mixture properties for air quality assessment in Mongolia
AU - Franklin, Meredith
AU - Chau, Khang
AU - Kalashnikova, Olga V.
AU - Garay, Michael J.
AU - Enebish, Temuulen
AU - Sorek-Hamer, Meytar
N1 - Publisher Copyright:
© 2019 by the authors.
PY - 2018/8/1
Y1 - 2018/8/1
N2 - Ulaanbaatar (UB), the capital city of Mongolia, has extremely poor wintertime air quality with fine particulate matter concentrations frequently exceeding 500 μg/m3, over 20 times the daily maximum guideline set by the World Health Organization. Intensive use of sulfur-rich coal for heating and cooking coupled with an atmospheric inversion amplified by the mid-continental Siberian anticyclone drive these high levels of air pollution. Ground-based air quality monitoring in Mongolia is sparse, making use of satellite observations of aerosol optical depth (AOD) instrumental for characterizing air pollution in the region. We harnessed data from the Multi-angle Imaging SpectroRadiometer (MISR) Version 23 (V23) aerosol product, which provides total column AOD and component-particle optical properties for 74 different aerosol mixtures at 4.4 km spatial resolution globally. To test the performance of the V23 product over Mongolia, we compared values of MISR AOD with spatially and temporally matched AOD from the Dalanzadgad AERONET site and find good agreement (correlation r = 0.845, and root-mean-square deviation RMSD = 0.071). Over UB, exploratory principal component analysis indicates that the 74 MISR AOD mixture profiles consisted primarily of small, spherical, non-absorbing aerosols in the wintertime, and contributions from medium and large dust particles in the summertime. Comparing several machine learning methods for relating the 74 MISR mixtures to ground-level pollutants, including particulate matter with aerodynamic diameters smaller than 2.5 μm (PM2.5) and 10 μm (PM10), as well as sulfur dioxide (SO2), a proxy for sulfate particles, we find that Support Vector Machine regression consistently has the highest predictive performance with median test R2 for PM2.5, PM10, and SO2 equal to 0.461, 0.063, and 0.508, respectively. These results indicate that the high-dimensional MISR AOD mixture set can provide reliable predictions of air pollution and can distinguish dominant particle types in the UB region.
AB - Ulaanbaatar (UB), the capital city of Mongolia, has extremely poor wintertime air quality with fine particulate matter concentrations frequently exceeding 500 μg/m3, over 20 times the daily maximum guideline set by the World Health Organization. Intensive use of sulfur-rich coal for heating and cooking coupled with an atmospheric inversion amplified by the mid-continental Siberian anticyclone drive these high levels of air pollution. Ground-based air quality monitoring in Mongolia is sparse, making use of satellite observations of aerosol optical depth (AOD) instrumental for characterizing air pollution in the region. We harnessed data from the Multi-angle Imaging SpectroRadiometer (MISR) Version 23 (V23) aerosol product, which provides total column AOD and component-particle optical properties for 74 different aerosol mixtures at 4.4 km spatial resolution globally. To test the performance of the V23 product over Mongolia, we compared values of MISR AOD with spatially and temporally matched AOD from the Dalanzadgad AERONET site and find good agreement (correlation r = 0.845, and root-mean-square deviation RMSD = 0.071). Over UB, exploratory principal component analysis indicates that the 74 MISR AOD mixture profiles consisted primarily of small, spherical, non-absorbing aerosols in the wintertime, and contributions from medium and large dust particles in the summertime. Comparing several machine learning methods for relating the 74 MISR mixtures to ground-level pollutants, including particulate matter with aerodynamic diameters smaller than 2.5 μm (PM2.5) and 10 μm (PM10), as well as sulfur dioxide (SO2), a proxy for sulfate particles, we find that Support Vector Machine regression consistently has the highest predictive performance with median test R2 for PM2.5, PM10, and SO2 equal to 0.461, 0.063, and 0.508, respectively. These results indicate that the high-dimensional MISR AOD mixture set can provide reliable predictions of air pollution and can distinguish dominant particle types in the UB region.
KW - Aerosol optical depth
KW - Aerosol types
KW - Air pollution
KW - MISR
KW - Machine learning
KW - Particulate matter
UR - http://www.scopus.com/inward/record.url?scp=85064136247&partnerID=8YFLogxK
U2 - 10.3390/RS10081317
DO - 10.3390/RS10081317
M3 - Article
AN - SCOPUS:85064136247
SN - 2072-4292
VL - 10
JO - Remote Sensing
JF - Remote Sensing
IS - 8
M1 - 1317
ER -