Presentation Information
[AAS08-P35]Satellite-to-Surface Mapping of Ammonia over South Korea by Merging CrIS and IASI Satellites with In-Situ Validation
*Chisung Yun1, Minho Kim1, MinSuk Bae2, Jung Min Park3, Yongjoo Choi1 (1.Hankuk University of Foreign Studies, Republic of Korea, 2.Mokpo National University, Republic of Korea, 3.National Institute of Environmental Research, Incheon, Republic of Korea)
Keywords:
Ammonia,Passive Sampling,Satellite(CrIS,IASI),Random Forest,Multiple linear Regression,Artificial Neural Network
Ammonia (NH3) is a key alkaline precursor that contributes to secondary inorganic aerosol formation through neutralization with sulfuric and nitric acids, thereby playing a significant role in PM2.5. The lack of in-situ NH3 measurements limits robust characterization of the spatio/temporal distribution and trends of NH3. In contrast, satellite observations provide large spatial coverage and long-term measurement but are still associated with high uncertainties. In this study, we developed a satellite-to-surface estimation framework by integrating surface NH3 concentration using nationwide passive-sampling, NH3 total columns retrieved from two satellites (IASI and CrIS), ERA5 meteorological variables, livestock density from Gridded Livestock of the World, MODIS land-cover fractions. Then we evaluated three empirical models including multiple linear regression (MLR), random forest (RF), and an artificial neural network (ANN) to estimate surface NH3 concentrations. Comparisons of NH3 between columns and in-situ showed weak correlation (R=0.21), whereas the performance of three models significantly improved in terms of statistical parameters. RF exhibited the highest cross-validated performance (R=0.87, RMSE=4.24 ppbv), followed by ANN (R=0.78, RMSE=5.26 ppbv) and MLR (R=0.51, RMSE=6.42 ppbv). The RF converged with the monthly variation of in-situ NH3, with monthly mean differences below 2 ppbv, both quantitatively and qualitatively, over the three-year period. These results demonstrate that the developed satellite-to-surface estimation framework can fill the gap between satellite retrievals and in-situ NH3, providing reliable information for large-scale, long-term monitoring in regions with sparse in-situ observations.