@article{jurczyski2026hybrid,
  title = {Hybrid Physics–Machine Learning Retrieval of AOD550 from HYPSO Hyperspectral Imagery},
  author = {Dawid JURCZYŃSKI},
  year = 2026,
  url = {https://ibimapublishing.com/p-articles/47AI/2026/4724326/},
  journal = {Communications of International Proceedings},
  volume = 2026 (12),
  abstract = {Atmospheric aerosols significantly influence air quality, climate processes, visibility, and environmental hazard assessment, making aerosol optical depth at 550 nm (AOD550) an important parameter in satellite-based monitoring systems. Existing aerosol retrieval approaches are often computationally expensive, dependent on complex radiative-transfer inversion methods, and difficult to apply in rapid-response workflows. At the same time, despite the growing availability of hyperspectral satellite systems such as HYPSO, relatively little research has focused on using HYPSO data for practical AOD550 retrieval in near-real-time environmental and threat-assessment applications. This study addresses this gap by proposing a hybrid physics–machine learning framework for rapid scene-level AOD550 estimation from HYPSO hyperspectral imagery. The methodology consists of two stages. First, a physics-inspired baseline estimate is generated using compact spectral, reflectance-derived, and geometric descriptors extracted from HYPSO observations. Second, a supervised machine-learning model is applied to estimate and correct the systematic residual bias of the baseline retrieval relative to a CAMS-based reference aerosol field. The framework was evaluated using Leave-One-Scene-Out cross-validation across 33 hyperspectral scenes. The results demonstrate a substantial improvement after the bias-correction stage, reducing the out-of-fold mean absolute error (MAE) from 0.234 to 0.024 and the root mean square error (RMSE) from 0.243 to 0.039. The findings indicate that HYPSO observations contain meaningful aerosol-related information and that hybrid physics–machine learning workflows can support rapid and computationally lightweight AOD550 estimation for satellite-data processing and environmental threat-assessment systems.},
  keywords = {HYPSO, aerosol optical depth, AOD550, hyperspectral imagery, machine learning, remote sensing, atmospheric aerosols, bias correction, satellite data processing, environmental threat assessment.},
  note = Article ID: 4724326
}