|
Fengyun-4A (FY-4A) is the second-generation geostationary orbit meteorological satellite series with higher observation frequency and resolution compared with the first-generation in China. While, the spatial resolution (4km) of the infrared channel and water vapor channel is lower than that of the visible light channel (1km), which limits the application of FY- 4A in extreme weather monitoring. At the same time, in order to adapt to the characteristics of the rapid time change of small and medium-scale meteorological disasters, this study based on the deep learning method to downscale the FY-4A satellite data in space and time. The approach consists of two main steps: first, FY-4A data is downscaled using a ESRGAN model transfer learning, which can extract spatially relevant information and reconstruct image resolutions such as infrared channels from 4km to 1km; second, based on the Super SloMo model, the time-related information can be extracted to effectively downscale the FY-4A data, and the temporal resolution of the FY-4A is reconstructed from 15min to 6min , making it comparable to the time resolution of weather radar. The spatial resolution evaluation based on the visible light channel shows that the method used in this study is superior to the spatial downscaling method of bicubic interpolation and Papoulis-Gerchberg in Peak Signal to Noise Ratio (PSNR), Structural Similarity (SSIM), Root Mean Square Error (RMSE), and Correlation Coefficient (CC), and can more effectively convert low-resolution FY-4A satellite data to the corresponding high-resolution satellite data. At the same time, the time-related information can be extracted based on the time downscaling model, the time resolution is converted from 15min to 6min, and the movement direction of the cloud remains the same. Compared with traditional methods, this downscaling approach is a postprocessing method of satellite data with higher precision, which can improve the application value of FY-4A in disaster weather warning.
|
