The training dataset is obtained by processing L1B data from AQUA/MODIS and GEO data. The data type is HDF file with a resolution of 1km. Each file contains six data tables: r (red channel reflectance, center wavelength 0.646 μ m), g (green channel reflectance, center wavelength 0.55 μ m), b (blue channel reflectance, center wavelength 0.469 μ m), lir (near-infrared reflectance, center wavelength 0.87 μ m), nir (shortwave near-infrared channel reflectance, center wavelength 1.64 μ m), flag (data quality indicator), latitude, and longitude. The reflectance measurement unit is db; The prediction dataset uses L1 level data from FY4A/AGRI. For the convenience of prediction, the data processing process is no longer separated separately, so there are no intermediate files. The data processing process is to first read the NOMHannel02 (red channel signal value, center wavelength 0.65 μ m), NOMHannel01 (blue channel signal value, center wavelength 0.47 μ m), NOMHannel03 (near-infrared signal value, center wavelength 0.83 μ m), and NOMHannel05 (short wave near-infrared channel signal value, center wavelength 1.61 μ m) of FY4A/AGRI L1 level data, and then read the corresponding CALChannel * (calibration values for each channel) of these four channels and convert the signal values into reflectance using the calibration values. The unit of signal value data is DN, the unit of reflectance data is db, and the calibration value is the conversion matrix from signal value to reflectance, without units. The predicted dataset contains data with four resolutions: 500M, 1000M, 2000M, and 4000M.

The AQUA/MODIS 1KM resolution L1B data and GEO data were obtained from NASA's official website（ https://modis.gsfc.nasa.gov/ ）Obtained data from the 32nd, 91st, 121st, 182nd, and 213rd days of 2019, totaling 5 days;

The FY4A/AGRI 1000M resolution L1 level data is provided by the National Satellite Meteorological Center. Includes data for the entire day of January 1, 2020.

   Training dataset: Read 5 channel reflectance data from the L1B dataset, read longitude and latitude from the GEO dataset, obtain the positions of all points in the above dataset that do not meet the threshold or are filled in values, generate an array with the same dimension as the above data table and all values of 0 as quality indicators, and change the values of all abnormal point positions to 1. Finally, store the processed 6 data tables in an intermediate file;

Validation dataset: Read 4 channel signal value data and 4 channel calibration value data, and use the value of each point in the signal value as the number of rows in the channel calibration value array to obtain the reflectivity corresponding to each point from the calibration value data. Due to the spectral differences between AQUA/MODIS and FY4A/AGRI instruments, it is necessary to perform spectral correction on the reflectance data of FY4A/AGRI. The instantaneous subsatellite point cross comparison method ("The Influence of Spectral Response Differences on High Precision Cross Calibration - Taking FY-3A/MERSI and EOS/MODIS as Research Examples") is used for spectral correction, which involves obtaining reflectance points of two instruments within a certain threshold in time and distance, calculating the magnification and difference of these points, and using these magnification and difference as correction coefficients to correct the reflectance of FY4A/AGRI. After calibration, input it into the model for prediction.