Downscaling Aerosol Optical Thickness from Satellite Observations: Physics and Machine Learning Approaches

In recent years, the satellite observation of aerosol properties has been greatly improved. As a result, the derivation of Aerosol Optical Thickness (AOT), one of the most popular atmospheric parameters used in air pollution monitoring, over ocean and continents from satellite observations shows comparable quality to ground-based measurements. Satellite AOT products is often applied for monitoring at global scale because of its coarse spatial resolution. However, monitoring at local scale such as over cities requires more detailed AOT information. The increase spatial resolution to suitable level has potential for applications of air pollution monitoring at global-to-local scale, detecting emission sources, deciding pollution management strategies, localizing aerosol estimation, etc. In this thesis, we investigated, proposed, implemented and validated algorithms to derive AOT maps with spatial resolution increased up to 1×1 km2 from MODerate resolution Imaging Spectrometer (MODIS) observations provided by National Aeronautics and Space Administration (NASA), while MODIS standard aerosol products provide maps at 10×10 km2 of spatial resolution. The solutions are considered on two perspectives: dynamical downscaling by improving the algorithm for remote sensing of tropospheric aerosol from MODIS and statistical downscaling using Support Vector Regression.