제목: AI-Transformed Air Quality Monitoring from Space​

​​​​​​​​​​​​​​​​​​연사: 김민석 박사 (UNIST)​​

​

일시: 2026년 7월 27일 월요일 16:00

장소: 과학관 B102호​​​

Abstract:

​The application of artificial intelligence (AI) has been rapidly expanding in the field of remote sensing. However, most AI-based remote sensing studies are data-driven and therefore tend to miss the physical processes involved in radiative transfer. Even when trained with highly accurate reference data such as ground-based measurements, these approaches cannot fully guarantee consistent performance across all spatiotemporal conditions. This study investigates AI applications in satellite-based air quality monitoring with the objective of retaining the advantages of AI while preserving physical interpretability. The first study introduces AI as a statistical tool for correcting errors in spectral aerosol optical depth (AOD), where a simple deep neural network model was trained to learn and correct wavelength-dependent errors in satellite AOD, and a spectral deconvolution algorithm was then used to physically retrieve the aerosol fine-mode fraction from the corrected spectral AODs. The second study replaces a radiative transfer model with an AI-based surrogate model to improve the computational speed and flexibility of a physics-based retrieval algorithm, enhancing the fusion of observations from two satellites and the retrieval of the imaginary refractive index at multiple wavelengths. Finally, an automated algorithm for methane plume detection was developed using AI-based image segmentation, where AI-based image processing addressed the limitations of geostationary satellites in terms of spatial resolution and sensitivity for methane plume detection.​