关键词: 土壤盐分, 回归克里格法, 支持向量机克里格法, 空间分布特征, 空间自相关, 山区积雪, 哨兵-2, Sen2Cor, 地形校正

Abstract: Snow cover is an important indicator of global climate change, and also an important freshwater resource in the arid regions of northwest China. Satellite remote sensing is an important way for snow cover identification and area change monitoring due to its large-scale, short-period, and repeated observations. Compared with Landsat TM, Sentinel2 image has advantages of higher radiation resolution, spectral resolution and spatial resolution, and higher time resolution, which undoubtedly is an important data source for snow research. At small and medium scales, the radiation brightness of highresolution satellite is affected not only by atmospheric attenuation such as atmospheric water vapor and aerosol, by topographical factors such as terrain shading, slope, aspect, but also by the surface bidirectional reflectance distribution function (BRDF), which distorts the surface information. Based on the digital elevation model (DEM), the atmosphere correction and topographic correction preprocessing were carried out synchronously on the basis of the surface BRDF characteristics using the mountain radiative transmission model, with 20 Sentinel-2 images of the middle section of Tianshan Mountain in China from May 2017 to April 2018. Then, snow cover information was accurately extracted by the normalization difference snow index NDSI and the snow reflectance threshold method in the green and nearinfrared bands. This method effectively eliminates the effects of water bodies, dense vegetation, shadows, and low radiance pixels on the recognition of snow. Moreover, combined with historical weather data, the variation characteristics of snow cover in the study area were analyzed within one year. Our results have great significance to the research on changes of surface reflectance and energy balance as well as terrestrial ecological environment caused by snow melting.

Key words: regression kriging, extreme learning machine kriging, spatial distribution characteristics, soil salinity, spatial autocorrelation, mountain snow cover, Sentinel-2, Sen2Cor, topographic correction.