GF-4/PMS与GF-1/WFV两种传感器地表反射率及NDVI一致性分析

2015年12月中国成功发射高分系列中首颗地球静止轨道卫星高分四号(GF-4),实现与高分一号(GF-1)近极地轨道卫星的优势互补,构成了具有多种空间和时间分辨率的对地观测体系。该文研究并分析了GF-4/PMS与GF-1/WFV地表反射率与NDVI的一致性,结果表明:一致性研究的最优空间尺度为50 m;GF-4/PMS与GF-1/WFV地表反射率存在较好的线性关系,各波段相关系数R均在0.7以上,传感器之间反射率的系统性偏差可以通过线性回归模型校正,校正后各波段反射率的RMSE明显降低;NDVI能够消除不同波段地表反射率"同增同减"偏差的影响,在GF-4地表反射率校正前后均表现出与GF-1较好的一致性,校正前后相关系数R分别为0.74和0.77。因此,GF-4在农业和植被遥感中具有较好的高分系列数据延续性和应用潜力。

Consistency analysis of surface reflectance and NDVI between GF-4/PMS and GF-1/WFV

Abstract: As the first geostationary satellite of China''s National High-resolution Observation Program (GaoFen Project), GaoFen-4 (GF-4) which launched in December 2015 enjoys the features of high temporal resolution and wide observation with a spatial resolution of 50 m. Meanwhile, GaoFen-4 together with GaoFen-1 (GF-1) polar-orbiting satellite forms a multi-spatial-temporal-resolution ground observation system, which provides more choices for agricultural remote sensing applications. Since the spectral responses of the Panchromatic Multispectral Sensor (PMS) boarded on GF-4 and Wide Field View (WFV) on GF-1 are slight different, there might be difference in reflectance measurements between them. To take fully advantage of their features, the imaging and vegetation detecting capability between two kinds of sensor are of great importance for us to explore in detail using image pairs from the areas of interest. In this study, we aimed at analysis of the continuity of surface reflectance and Normalized Difference Vegetation Index (NDVI) based on GF-4/PMS and GF-1/WFV sensors images. Two image pairs of Northern China Plain and its nearby mountain areas of China, acquired in May 2016, were employed for the comparison. Radiometric calibration, atmospheric correction and geometric correction were operated at first on the Level 1A images. Since there is discrepancy in spatial resolution between GF-4/PMS (50 m) and GF-1/WFV (15 m) images, two pixel-resampling methods were adopted not only to get a comparable surface reflectance image pair, but also to explore the scale effect on their continuity. Statistic indicators such as R2 and RMSE were used to measure the continuity of their surface reflectance and NDVI, and linear regression model was adopted for surface reflectance rectification of GF-4/PMS images. Our results demonstrated that with spatial scale of the pixels increasing, the correlation of surface reflectance between GF-4 and GF-1 is also in increment. However, the uptrend of correlation speeds down or even reverses in blue band from 50 to 80 m. So 50 m is considered as an optimal spatial scale for the continuity study of GF-4 and GF-1 surface reflectance. The surface reflectance of GF-4 and GF-1 has a high positive correlation with R greater than 0.7 in all bands, but there is also bias. The bias includes systematic bias and random bias caused by sensors differences and outside factors, respectively. The major influence factor is the sensors'' spectral response function, and it affects red band least. Other outside influence factors may include acquiring time, variation in atmospheric conditions and solar azimuths. Besides, BRDF effects were not under consideration in this paper either. The systematic bias can be reliably modified by a linear regression model, thus to predict GF-4/PMS reflectance from GF-1/WFV reflectance with an elimination of RMSE on surface reflectance between GF-4 and GF-1 remarkably. Meanwhile, the correlation coefficient R of NDVI increased slightly from 0.74 to 0.77 by the process of surface reflectance rectification, which proves that NDVI could weaken the effect of the surface reflectance bias. As a result, NDVI represents a much better continuity between GF-4 and GF-1 both before and after GF-4 surface reflectance rectification. Based on our experiment results, we suggest that GF-4 has a good continuity of China''s GaoFen series satellites that enable its potential applications in the monitoring of agriculture and vegetation.