关键词: 积雪反射率光谱库, 光谱混合分析, 积雪面积, 雪粒径, 渐进辐射传输模型, 端元变化

Abstract:

Snow-cover information is important for a wide variety of scientific studies, water supply and management applications. The NASA Earth Observing System (EOS) Moderate Resolution Imaging Spectroradiometer (MODIS) provides improved capabilities of observing snow cover from space and has been successfully using a normalized difference snow index (NDSI), along with threshold tests, to provide global automated binary maps of snow cover. NDSI and other classification algorithms were used to inverse subpixel information of Snow Cover Fraction (SCF), but these algorithms neglected the relation between SCF and Snow Grain Size (SGZ). The SGZ might affect snow reflectance spectral curves, while most subpixel classification algorithm took advantage of the spectral feature space. The collaborative inversion of SCF and SGZ helped improve the understanding of the physical properties of snow. Meanwhile, it was possible to improve the retrieval accuracy of SCF. The framework of spectral mixture analysis (SMA) was widely used in the target detection of remote sensing images because of its ability to extract subpixel information and SMA could use mathematical methods to model SCF with snow reflectance spectral curves with different snow grain sizes. In this paper, in view of the snow cover with MODIS remote sensing image, based on the framework of spectral mixture analysis, the snow reflectance spectral library with different grain sizes was built by asymptotic radiative transfer (ART) model, and a sparse unmixing algorithm of snow cover fraction retrieval was proposed considering the endmember variability of snow with other materials and bilinear radiative process of endmembers. The ART model had a higher efficiency compared with MIE scatter model. Meanwhile, ART model considered snow grain shape parameters. The majority algorithm of SMA assumed the endmembers independent, which might neglect the interaction of endmembers, while bilinear radiative process of endmembers could consider second-order scattering effects, which had physical meaning. This algorithm firstly used asymptotic radiative transfer model to establish reflectance spectral library with different grain sizes, and Sequential Maximum Angle Convex Cone (SMACC) endmember extraction algorithm was used to obtain the spectral library of vegetation, soil and rock shadow. After the establishment of a variety of spectral libraries, the root mean square error index was used to get the optimal combination of endmembers for each pixel as MODIS Snow Covered Area and Grain Sizes(MODSCAG) model, which could accurately describe the endmember variability. After the optimal endmembers combination obtained, the bilinear radiative process was added into sparse regression spectral mixture analysis to simultaneously obtain snow cover fraction and snow grain size. Experimental results showed that this method could simultaneously inverse the snow grain sizes and snow cover fraction, and the retrieved snow grain sizes is smaller than that from single band of asymptotic radiative transfer model The accuracy of retrieved snow cover fraction is increased slightly compared with MOD10A1 product.

Key words: Snow Reflectance Spectral Library, Spectral Mixture Analysis, Snow Grain Size, Snow Cover Fraction, Asymptotic Radiative Transfer Model, Endmembers Variability