基于渐进辐射传输模型的雪粒径与积雪覆盖面积反演算法

Snow Grain and Snow Fraction Retrieval Algorithms based on Asymptotic Radiative Transfer Model

渐进辐射传输模型(Asymptotic Radiative Transfer,ART)广泛应用于雪粒径反演,但是ART模型忽略了像元混合的影响。在本文中,ART模型被应用于混合像元分解之中,考虑积雪粒径变化的影响,提出了两种混合像元分解算法,用以同时获取雪粒径和积雪面积。第一种算法是采用均方根误差指标获取每个像元的最优雪粒径,使用全约束线性分解算法(FCLS)反演积雪面积。另一种算法使用通用梯度下降算法(GRG)获取每个像元的最优粒径,仍然采用全约束线性分解算法获取积雪面积。为了提高模型的运行效率,单波段ART算法用来获取影像的雪粒径,将之作为两种算法的先验知识。同时,MODSCAG模型反演的结果也被用于进行交叉对比,实验结果表明,两种算法比MODSCAG模型的反演结果更为精确。

Asymptotic Radiative Transfer （ART） model had been widely applied to pure snow grains sizes retrieval,but the model neglected the effect of mixed pixels.In the study we extended ART model to mixed pixels.In order to obtain snow grain and snow cover fraction,two kind unmixing algorithms were used.One algorithm used the non-constrained least square method of MODSCAG model to obtain snow grain sizes for each pixel,and Full Constrained Least Square （FCLS） unmixing algorithm was used to reverse the snow cover fraction for each pixel.The other algorithm used Generalized Reduced Gradient （GRG） algorithm to decompose each pixel,and snow grain sizes and snow cover fraction were obtained simultaneously.Before unmixing image, ART model was used to build a snow reflectance spectral library with different snow grain sizes,meanwhile,vegetation library （selected from image） and soil library （selected from Hopkins univer- sity spectral library） were used.In order to compare the retrieval accuracy for snow grain sizes and snow cover fraction, single channel ART snow grains retrieval algorithm （which is called as Zege model） and MODSCAG model were used.The experimental results showed that the retrieval results of two algorithms had higher accuracy than MODSCAG model.