Abstract: As one of the most important surface water resources in arid areas,snowcover plays an irreplaceable role in water resources management,global climate change,snowmelt runoff simulation and forecasting. Snow porosity is one of the important indexes reflecting physical characterization of snowmelt and snowpack. Using the hyper spectral technology,this paper analyzed the impacts of snow porosity on spectral reflectance in a typical snowmelt watershed in Northern Slope of the Tianshan Mountain. Combined the BP neural network method and partial least-squares regression,we further proposed a new quantitative modeling approach of hyper spectral to specifically invert the snow porosity by remote sensing data. Results show as follows:(1)the snow porosity was obviously responded to spectral reflectance,and showed significant correlation in the near-infrared bands including 1165-1478 nm,1615-1921 nm,and 1930-2100 nm.(2)When the implied nodes was set 3,the inversion ability of snow porosity with hyper spectral data can be greatly improved by PLS-BP model,coefficient of determination R² and the root mean square error(RMSE) of linear regression between the snow porosity measured and simulated of PLS-BP in the same period were 0.9159 and 0.04,respectively.(3)Compared with the traditional partial least-squares regression(PLSR) and principal components regression(PCR),PLS-BP model obtained higher inversion,for example,the coefficient of determination R2 and the root mean square error (RMSE) of PCA-BP were 0.6042 and 0.42,respectively,but for the traditional PLSR,they were 0.3536 and 0.63,respectively. The innovation of this study is as follows:(1)using principal component to obtain main information bands as input of PLS-BP model,which is conducive to reflect reflectivity of snow porosity and help improving the model precision;(2)the PLS-BP combined with hyperspectral remote sensing data could estimate the snow porosity better,and this method can be referenced for analysis of near infrared spectrum;(3)the monitoring and estimating model can not only predict snow indexes but also can provide remote assistance for snowmelt flow and flood forecasting.

Key words: hyperspectral remote sensing, melt period, snow porosity, PLS-BP