Abstract:
Abstract: GF-1 satellite is the first satellite of the high resolution satellite series in China. Since its successful launch on April 26 2013, GF-1 satellite has been widely applied in agricultural remote sensing monitoring practice in China, and it has become a major data source of agricultural remote sensing dynamic monitoring. Based on the principle of radioactive transfer model of 6S (second simulation of a satellite signal in the solar spectrum), this paper designed and realized the algorithm and program suitable for GF-1 satellite data atmospheric correction. By using the 6S model, the algorithm obtains the parameters for the conversion from reflectivity (or irradiance) of Top Of Atmosphere (TOA) to surface reflectance, and then calculates the surface reflectance of each pixel of each image according to the conversion parameter. The algorithm takes GF-1 satellite first level data, metadata, and open parameter of sensor as the input data, without auxiliary data from other sources. The specific process includes 3 steps, i.e. radiometric calibration, running parameters settings and atmospheric correction. Radiometric calibration is to convert the DN (digital number) value of the original GF-1 satellite first level image into radiation brightness, and then calculate apparent reflectance by combining the reflectivity (or irradiance) of TOA. Either reflectivity (or irradiance) of TOA or apparent reflectance can be taken as the input of atmospheric correction program. Precondition for realizing the algorithm is to calculate the average solar irradiance parameters of each wave band of satellite sensor atmospheric top according to spectral response function of GF-1 satellite sensor and WRC (world radiation center) sun spectrum function. Operation parameters include 2 types: 1) input of satellite images, including satellite zenith angle, satellite azimuth angle, solar zenith angle, solar azimuth, sensor height, ground elevation, radiation calibration coefficient and spectral response functions of various loads, which can be acquired from the metadata of the images; 2) atmospheric model parameters, such as atmospheric model, atmospheric aerosol model, visibility, solar spectrum function. The default value will be set by the system according to the data conditions, and it can be adjusted according to the real situation. Spectral response function of GF-1 satellite is provided by the satellite producer, and the re-sampling is the spectral response curve with the resolution of 2.5 nm and it is input into the 6S model. Atmospheric correction is to convert the apparent reflectance image (or radiation brightness) into ground reflectance. Now, the input is the GF-1 apparent reflectance image (or radiation brightness) which needs atmospheric correction and the output is the ground reflectance image. On the basis of the development of the algorithm, the Fortran and interface description language are applied to compile atmospheric correction batch processing programs, so as to realize the batch processing during atmospheric correction process. This paper used the data of GF1 WFV (wide field view) of Beijing region on April 3, June 28, and November 2, 2014, and January 19, 2015, which 4 phases represented 4 seasons, i.e. spring, summer, autumn and winter. By using the atmospheric correction result of FLAASH (fast line-of-sight atmospheric analysis of spectral hypercubes) of ENVI software, the evaluation was conducted. The relative deviation of the whole year for 4 phases between the 2 methods was 3.26%. Blue light band had the highest deviation of 11.21%, followed by red, near-infrared, and green light bands, which were 1.19%, 0.73% and 0.24% respectively. The average relative error in the areas covered by crops was 12.99%, the highest was in winter which was 17.40% and those in autumn and spring were 15.02% and 14.15% respectively, and summer had the lowest value of 8.31%. Whole correction of ground reflectance of various bands didn't show significant difference, but the reflectance of various bands after 6S correction was usually slightly higher than the correction result of FLAASH. The calculation results of the NDVI (normalized difference vegetation index) based on 2 correction results were basically same with the relative deviation of 0.64%, and the absolute difference was within 0.0548 except water body. In terms of calculation efficiency, the 6S model has realized the batch calculation which was not provided in the commercial software of FLAASH module. Under the same hardware environment, the calculation efficiency was improved by more than 75.0%. The research result shows that the atmospheric correction program developed by this paper can stably process GF-1 satellite data in batch, and it can be used as a component of agricultural remote sensing monitoring operation.