Research On Snow Parameters In Northeastern China Based On Radar Data

The use of remote sensing data for snow measurement has gradually become the mainstream method.Radar can observe the earth day and night without being affected by clouds and sunlight,and provide images with high spatial resolution.These advantages effectively make up for the deficiency of spectral data and passive microwave data.This paper mainly used Sentinel-1 SAR data to research the snow parameters in Northeastern China.The specific research work and innovations are as follows:(1)Snow Covered Area Estimation for Forest Area in Northeastern China Based on Multi-temporal Sentinel-1 SAR Data.It is difficult for researchers to identify the snow coverage in forest areas from spectral images.In this paper,the field measurement of forest parameters in Jingyuetan Forest Park was carried out.The measured parameters and sentinel-1SAR data were used as the input data of the forest scattering model.Then,the backscatter coefficient after eliminating the influence of the forest canopy was calculated.Finally,the snow coverage of study area was calculated and the distribution characteristics of snow coverage in different forest areas were analyzed.(2)Snow Water Equivalent Retrieval Algorithm in Jilin Province of China Based on Multi-Temporal Sentinel-1 Data.Through experimental analysis,this paper found the relationship between the measured snow parameters in the study area and the backscatter coefficient difference of the autumn and winter SAR data,and established an equation.Then the backscatter coefficient difference was used to calculate the SWE instead of the snow thermal resistance.The experimental result has a MAE of 0.198 cm,and the RMAE is 0.240 cm.The retrieval accuracy is high.(3)Snow Depth Retrieval in Northeastern China Based on Machine Learning and Sentinel-1 SAR Data.Machine learning algorithms(MLAs)does not need to consider the physical model and is gradually applied to the research of SD.However,MLAs are susceptible to input parameters.Therefore,how to select suitable parameters according to the condition of snow underlying surface is a question worthy of discussion.Based on the snow data of the field measurement and meteorology station in Northeast China,using C-band data with 20 m spatial resolution,this paper proposed two parameter selection methods,including the correlation coefficient method and the machine learning fusion method,to discuss the influence of different parameter combination(PC)on SD estimation.Then XGBoost,Random Forest,Linear Support Vector Regression and kernel Support Vector Regression were used to estimate SD based on the selection input parameters,and the influence of different MLAs on SD retrieval was discussed.The experimental results demonstrated that the selected PC by the correlation coefficient method and XGBoost algorithm achieved best SD results in the study area.In cropland areas,the average value of MAE and RMSE were 1.75 cm and 2.58 cm,respectively.In forest areas,the average value of MAE and RMSE were 3.12 cm and 5.07 cm,respectively.In this article,Sentinel-1 SAR data and measured data were used as the main data sources.The semi-empirical model and machine learning method were used to estimate the snow cover area for forest,farmland snow water equivalent and large-scale snow depth in northeastern China.