融雪期积雪深度高光谱反演方法研究

积雪在干旱区的水分平衡中发挥着极为重要的作用，积雪深度的监测主要依靠地面站点观测和遥感反演等技术，高光谱遥感为快速、大面积监测积雪的物理特性提供了可能。通过对融雪期不同厚度积雪表面的反射光谱以及积雪深度数据的观测，进而对二者进行相关性分析;采用相关性较高同时也是特征吸收谷的波段数据建立单波段雪深回归模型;采用呈显著相关的波段进行逐步回归，选用贡献率最高的波段作为神经网络模型的输入变量进行积雪深度的反演研究。结果表明：在天山北坡中段的军塘湖流域地区，1 022，1 241和1 492 nm附近是积雪的特征吸收谷;相比单波段反演雪深模型的估算精度(R2=0.53)，BP神经网络模型具有更高的雪深反演水平，当隐含层节点数为4时，R2为0.86，RMSE为0.67，表明神经网络模型可以显著提高高光谱数据反演积雪深度的能力。

Retrieval Method for Estimating Snow Depth Using Hyperspectral Data in Snowmelt Period

The snow surface reflectance spectra with different depth in snowmelt period and snow depth data were measured and its correlation was analyzed. The characteristic absorption band data of the valley with higher correlation were used to establish a single band snow depth regression model. The highest contribution rate of the band was selected as the input variable of the neural network model to retrieve snow depth. The results show that in Juntang Lake area, near 1 022, 1 241 and 1 492 nm exists characteristic absorption valley of snow, and compared to estimation accuracy of the single-band inversion of snow depth model (R2=0.53), ANN-BP model has a higher inversion level, and determination coefficient (R2=0.86, RMSE=0.67) was obtained with 4 nodes in hidden layers, indicating that ANN-BP model can greatly improve the ability of inversion of snow depth with hyperspectral data.