Transfer learning-based retrieval of cloud base height from FY-4B satellite observations

Cloud base height (CBH) is a key parameter for cloud-aerosol-precipitation interactions and aviation meteorological support. This paper proposes a hybrid algorithm based on deep learning and transfer learning, which integrates level one radiative data from the geostationary satellite FY-4B/AGRI with CBH products from the polar-orbiting satellite NPP/VIIRS. A deep neural-network baseline model is first established and then refined via transfer learning using EarthCARE/CPR observational results as true value, enabling CBH retrieval from geostationary data. The result shows that the RMSE for CBH during the day and night are 1.99km and 2.48 km, respectively, which is comparable to the CBH product accuracy of NPP/VIIRS, which provides a new technical approach for obtaining all-weather, large-scale CBH measurements.