Retrieval Of Seasonal Snow Depth In Juntanghu Basin Of Tianshan Mountains Based On Sentinel-1 Data

Snow cover is one of the important water resources in arid and semi-arid regions,and seasonal snow melt is great important to agricultural production and ecological development in the coming year.The snow depth is an important parameter for the calculation of snow water equivalent,and provides important data support for the early warning of spring flood.This article selects Juntanghu River Basin,in Hutubi county,Changji,Xinjiang,as the study area.Selected Sentinel-1 data with IW mode Level 1 SLC of2016815-2017223 and 2016815-2017319 two image pairs as the basis data.field measured snow depth data and ultrasonic snow depth data of February 23,2017 to March 2017,28 as test data.When collecting test data,generally choose points which have the wide view,typical terrain and less interference,and measured the actual snow depth,environmental temperature,moisture content and density data of snow.At the same time,set up ultrasonic snow sounder and weather station at high altitude and complicated terrain,so that the data can be collected for a long time.Study used two rail differential interferometry synthetic aperture radar technology(D-InSAR)method through selecting image pairs,computing baseline,flat-earth phase removal,filter noise reduction,phase unwrapping and geocoding to calculating Snow phase of study area.Then using kriging interpolation method completes translating point data to the surface data of the study area,at the same time,using geostatistics verified the snow depth inversion results and qualitative analyzed inversion result of the snow depth.Using BP neural network model with Inversion of snow depth,Snow density and moisture content of snow cover to improve the accuracy of inversion.The main conclusions of this study are as follows:1.The D-InSAR result of inversion for snow depth value range on February 23,2017 is 12.455 cm-12.455 cm,mainly snow in 15 to 30 cm deep.Snow depth on March 19,2017,the performance range in 10.43 cm-33.26 cm,the main scope of snow depth in 12-21 cm.Select the measured data at 12:10 a.m.each day,linear regression analysis was performed with inversion results.The correlation coefficients of some sites are 0.61 and 0.66,due to the mountain cover and complicated terrain.2.Interpolation inversion.All point based on the measured snow depth(including 8 sharing point data of Hongshan reservoir management stations)and ultrasonic data to kriging interpolation,snow depth data value range on February 23,2017 is in 9.26cm-39.99cm,March 19,the interpolation result is 8.55 cm to 32.93cm.Analysis different snow depth of inversion and interpolation,and found the interpolation result is overall lower than the inversion of that,relatively speaking,the inversion results is closer to the actual snow depth value,but the snow depth overall trends are the same,which are increasing from the northeast to the southwest of snow depth gradually.To qualitative analysis the result of Snow depth,It is found that the inversion results are good and suitable for the inversion of snow depth in the research area.3.The BP neural network.In order to improve the precision of inversion results,this study integrated multi-factor jointly building BP neural network model,and the result of the deep snow inversion is better.BP neural network improve the fitting degree of the small sample data,and quick to build the model.by using the model integrated data of the Sentinel-1,environmental temperature,snow density and moisture content of snow to construct the snow depth inversion model.Running the BP neural network model by using random 11 points data to construct the model,getting snow depth model precision R~2=0.84 of February 23,and R~2=0.88 of March19.Getting accuracy R~2=0.77 of precision by random 5 sites measured data inspection on 23 February,R~2=0.81 on March 19.It is found from the results:using the BP neural network can effectively support remote sensing inversion,and obviously improve the precision of the inversion results.To sum up,in this paper,through D-InSAR technology to retrieve snow depth,acquiring microwave inversion data,then using BP neural network integrated a number of factors to build the BP neural network model of snow depth inversion,getting better results.In the future,it will help for the inverting seasonal snow depth and snow melt water,and it has a certain practical significance and value.