Research On NDSI Multi-threshold Mountain Snow Cover Extraction Based On Sentinel-2 Imagery

Snow cover is one of the most active elements of the cryosphere and plays an important role in the surface radiation budget and water balance.The Sentinel-2 A/B satellite has high spatial and temporal resolution,radiometric resolution and spectral resolution,especially its high radiometric resolution reduces the problem of snow/ice saturation in remote sensing observations,and has become an important data source for snow cover extraction.However,in mountainous areas with complex terrain,snow cover extraction has the following difficulties:1)The topographic correction effect of the surface reflectance L2 A product provided by the European Space Agency(ESA)is not ideal,making it difficult to identify snow cover in the shadow of the topographic;2)The normalized differential snow cover index(NDSI)and the normalized differential forest snow cover index(NDFSI)are important methods for snow cover extraction,but due to the strong spatial heterogeneity and fast change speed of snow cover in mountainous areas,Using the classical fixed threshold method to extract snow cover will result in a large misclassification error.Therefore,this paper takes the Babao River Basin in the northeastern Qilian mountain as the research area,and firstly uses the Sen2 Cor topographic normalization tool provided by ESA to re-synchronize the atmosphere and topographic of Sentinel-2 L1 C products by configuring auxiliary data and activating topographic correction parameters,that is,topographic normalization;Then,in order to improve the accuracy of snow cover extraction in mountainous areas,NDSI/NDFSI multi-threshold were constructed by integrating snow cover period type,aspect and snow underlying land coverage type;Finally,by testing the optimal values of multi-threshold,the dynamic change characteristics of snow cover in a snow cover period(2020/10-2021/05)are analyzed.The main research contents and conclusions are as follows:(1)Using the ESA Sen2 Cor topographic normalization tool,a user-defined L2 A product(S2C-L2A)was obtained by activating cirrus correction and BRDF correction,introducing 12.5 m DEM and scene classification ESACCI-LC auxiliary data.Compared with the ESA-L2 A product,the S2C-L2 A product has a better effect of eliminating the atmosphere and topographic,especially improving the ability to identify snow cover in the shadow.(2)The effects of snow cover period,aspect and snow underlying land coverage type on the snow cover NDSI/NDFSI threshold were comprehensively analyzed.The NDSI/NDFSI thresholds of grassland sunny aspect(GSU),grassland shady aspect(GSH),woodland sunny aspect(WSU),woodland shady aspect(WSH),other sunny aspect(OSU),other shady aspect(OSH)and flat were determined respectively.Compared with the high-resolution GF-2 snow map results,it is found that the multi-threshold method has higher accuracy in extracting snow cover,and the overall classification accuracy,omission error,commission error and Kappa coefficient are 97.70%,0.20%,11.17% and 0.93 respectively.At the same time,compared with the snow cover results extracted by the fixed threshold,the multithreshold reduces the influence of the aspect and the snow underlying land coverage type on the NDSI/NDFSI threshold,and the overall classification accuracy and Kappa coefficient are improved,omission error and commission error are reduced.In addition,compared with the Sen2 Cor snow cover extraction algorithm,the snow cover extraction accuracy of the multi-threshold method is higher.(3)During a snow cover period,the spatial distribution of snow cover in the Babao River Basin varies greatly.During the accumulation period,the snow cover is only distributed in the mountainous area,during the stable period,the snow cover is distributed in the high-altitude mountains and river valleys,during the ablation period,the snow cover is only scattered in the high-altitude mountains.The snow cover rate of the Basin fluctuated greatly with time,the snow cover rate firstly increased and then decreased during the accumulation period,stabilized at 15%-20%during the stable period,and during the ablation period,the snow cover rate firstly decreased,then increased and then decreased during the ablation period.