Xiao, M. J., D. S. Fu, H. R. Shi, G. C. Wang, H. C. Lei, X. L. Han, and X.-A. Xia, 2025: Evaluating and enhancing Fengyun AGRI cloud detection with the ensemble learning algorithm. Adv. Atmos. Sci., https://doi.org/10.1007/s00376-024-4206-7.
Citation: Xiao, M. J., D. S. Fu, H. R. Shi, G. C. Wang, H. C. Lei, X. L. Han, and X.-A. Xia, 2025: Evaluating and enhancing Fengyun AGRI cloud detection with the ensemble learning algorithm. Adv. Atmos. Sci., https://doi.org/10.1007/s00376-024-4206-7.

Evaluating and Enhancing Fengyun AGRI Cloud Detection with the Ensemble Learning Algorithm

  • Cloud detection is critical for satellite remote sensing algorithms and downstream applications. This study evaluates the official cloud mask (CLM) product of the Advanced Geostationary Radiation Imager (AGRI) on the Fengyun-4A (FY-4A) satellite using two years of data from the Cloud Aerosol Lidar with Orthogonal Polarization (CALIOP) over China. The evaluation reveals moderate performance, with an accuracy (ACC) of 79% for daytime and 80% for nighttime. However, the FY-4A CLM struggles with high false-positive rates (FPRs of 35% during the day and 23% at night), often misclassifying clear skies as clouds. To address this issue, a random forest (RF)-based cloud detection algorithm is developed, using collocated CALIOP observations as reference labels for model development and validation. The models are divided into daytime and nighttime categories based on the solar zenith angle. Feature engineering demonstrates that adding temporal information, spatial texture information, and dynamic surface features reduces FPR values significantly. The ACC of the daytime (nighttime) model improved by up to 13.6% (10.2%). The proposed RF models achieve exceptional cloud detection, with ACC and true-positive rates (TPR) exceeding 90% with an FPR below 10% for both day and night, outperforming the FY-4A CLM. Compared to MODIS and FY-4A CLM, the RF-based models demonstrate superior accuracy in identifying clouds under challenging conditions such as dust, snow, and high pollution. This study offers a promising alternative to enhance cloud detection for the FY-4A imager.
  • loading

Catalog

    Turn off MathJax
    Article Contents

    /

    DownLoad:  Full-Size Img  PowerPoint
    Return
    Return