利用光谱混合分解模型分析GF-6新增波段对土地利用/覆被的响应

当前面对紧迫的自然资源管理压力和生态环境监测需求,针对国产遥感卫星大数据应用能力的挖掘将面临很大的挑战。GF-6卫星具有大角度、高频次和新谱段的特点,该文基于GF-6卫星数据,测试新增的红边、黄光和紫光波段响应能力。利用具有物理意义的全约束线性光谱混合分解模型,根据研究区物候特征确定四端元包括植被(GV),裸地和建设用地等基质(SU),山体植被阴影(DA)以及水(WA),通过对比保留红边、黄光波段、紫光波段和去除红边、黄光、紫光波段后的分解结果,对各新增波段和GV端元、SU端元、差均方根(RMSE)进行相关性分析;最后对比光谱混合分解结果和基于专家知识决策树分类结果。通过对比丰度值估计参数和决策树分类结果发现红边波段对植被较为敏感,对光谱混合分解模型的适用性、稳健性以及丰度值估计精度有着很大贡献,黄光波段和紫光波段经过数据降维后对植被和裸地、建设用地有少量贡献。通过相关性分析发现红边2波段、近红外波段与GV端元丰度图有最大的相关性,紫光波段、黄光波段和红边1波段与GV端元反向相关;红边1波段、紫光波段和黄光波段与SU端元丰度图显著相关;红边1波段和黄光波段对丰度值计算误差有主要贡献,是主要的噪音来源,紫光波段次之。通过对比GF-6数据和OLI、Sentinel-2数据丰度值估计结果发现GF-6丰度值估计的均方根误差以及除了WA端元的各端元丰度值估计变异系数均小于OLI和Sentinel-2载荷,体现出CF-6卫星在地表信息识别上较高的精度和稳健性。

Response of new bands in GF-6 to land use/cover based on linear spectral mixture analysis model

The pressure of natural resource management and ecological environment monitoring is increasingly prominent. It is urgent to give full play to the advantages of remote sensing data to assist the natural resources management. The application capability of domestic satellites needs further excavation. GF-6 is a newly launched satellite belonging to China High Resolution Earth Observation System, which has the advantages of large angle, high frequency and new spectrum. CF-6 is one of the few satellites with eight bands in the visible and near-infrared spectrum. A control experiment was designed for the test of new bands of GF-6 based on methods of linear spectral mixing analysis (LSMA) model, decision tree and correlation analysis. The complete spectral space was reconstructed into four scenarios: the original spectral space (S1), the lack of red-edge band scenario (S2), the lack of yellow-band scenario (S3) and the lack of purple-band scenario (S4). All the research work was based on endmember (EM) fraction maps, which were generated from LSMA. In order to obtain the endmember fraction maps accurately, we employed the principal component analysis (PCA) to reduce the data dimensions, and determined four endmembers (Green vegetation, GV; Substrate, SU; Dark material, DA and Water, WA) though the result of PCA and the status of local Land use/cover. After that, the contribution of new bands to endmember fraction maps was judged by correlation analysis between the add-bands and each endmember fraction maps. Finally, the decision tree classification was used to observe the classification results in scenarios and draw the final conclusion. In addition, we also compared the application ability of GF-6 with OLI and Sentinel-2 with LSMA model. Through the four situation''s experiments, we came to conclusions as follow. The results of all three methods show that the red-edge band is sensitive to vegetation, which can effectively improve the recognition accuracy of vegetation. Besides, the result of LSMA model indicates that the red-edge band also contributes to the applicability and stability of the LSMA model. The result of correlation analysis shows that violet band and yellow band have strong correlation with substrate, therefore contributing to the classification of urban interior facilities; but they have an inverse correlation with vegetation. The yellow band and red-edge1 band may cause classification errors in mapping of large area. The significance of this study lies in it founds a variety of response characteristics of new bands in GF-6 to land use/cover. And the conclusion of this study will not only provide a robust support for natural resources supervision and ecological protection in our country, but also witness the great progress made by Chinese satellites.