基于6S模型的GF-1卫星影像大气校正及效果

GF-1 satellite image atmospheric correction based on 6S model and its effect

高分一号(GF-1)卫星是中国高分系列卫星的首发星,自2013年4月成功发射以来,在中国农业遥感业务工作中得到了广泛应用,已成为中国大宗农作物种植面积遥感监测的主要数据源。该文基于6S(second simulation of a satellite signal in the solar spectrum)辐射传输模型原理,设计并实现了适合于GF-1卫星数据大气校正算法与程序。算法以GF-1卫星1级数据、元数据及传感器公开参数为输入数据,不需要其他外源辅助数据,经过辐射定标,计算各波段平均太阳辐射值、表观反射率,通过选择大气模式,驱动6S模型获取表观反射率转换为地表反射率的参数,逐像元计算影像地表反射率。在算法研制的基础上,应用Fortran和IDL语言编写了大气校正批处理程序,实现了大气校正过程的批处理。该文采用2014年4月3日、6月28日、11月2日,以及2015年1月19日4个时相北京地区GF1卫星WFV(wide field view)数据,分别代表春夏秋冬4个季节,通过与ENVI软件的FLAASH(fast line-of-sight atmospheric analysis of spectral hypercubes)大气校正结果对比进行评估。2种方法 4个时相各波段全年相对偏差为3.26%,蓝光波段偏差最大为11.21%,其次是红、近红和绿光波段,分别为1.19%、0.73%和0.24%。作物覆盖区平均相对误差为12.99%,冬季最高为17.40%,秋季和春季分别为15.02%和14.15%,夏季相对差异最小为8.31%。各波段地表反射率的整体校正情况并未有太大差异,但6S校正后各波段反射率普遍比FLAASH校正结果略微偏高。2种校正结果计算的NDVI也基本一致,相对偏差0.64%;除水体外,绝对值差值的平均值均在0.0548以内。从计算效率来分析,6S模块实现了商用软件FLAASH模块中未提供的批量计算,在相同硬件环境下计算效率提高了75.0%以上。研究结果表明了该文开发的大气校正程序能够稳定批量处理GF-1卫星数据,可以作为农业遥感监测业务流程的组成部分。

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.