Study On The Snow Depth Spatial-temporal Variation Characteristic In The Northern Hemisphere

Snow is one of the most important components of the cryosphere,and is the most sensitive natural element for weather and climate response.It affects local or regional water resources and energy balance,hydrological processes,and ecosystem functions.Passive microwave satellite remote sensing can penetrate the clouds and the atmosphere,and has the characteristics of working in all-weather and all-day.This makes using passive microwave satellite remote sensing data to estimate snow cover depth,snow water equivalent,and other snow parameters have greater advantages.In recent years,the research on remote sensing of snow depth and snow water equivalent by using passive microwave satellites has made great progress.In this paper,the research on the inversion of snow depth using passive microwave remote sensing data and ground measurement data is carried out,and a relatively simple snow depth estimation method is attempted.This method was used to invert the snow depth in the northern hemisphere,and generated a daily snow depth data set for the northern hemisphere in the past 25 years.Finally,the temporal and spatial characteristics of snow cover in the Northern Hemisphere were analyzed over the past 25 years.This paper firstly proposes a new snow depth retrieval algorithm based on the research of predecessors' snow depth inversion algorithms study,considering the evolution of snow properties and the heterogeneity of snow properties on the spatial and temporal scales.This algorithm takes full account of the impact of vegetation type on the underlying surface on the properties of snow cover,as well as site location information and snow cover time information.Finally,a snow depth inversion algorithm based support vector regression method was established.Firstly,the snow depth of the Eurasian continent was retrieved using this algorithm,and then the ability to estimate snow depth was compared with the other four existing snow depth retreiveal algorithms.The other four algorithms include: Chang algorithm,Spectral Polarization Difference(SPD)algorithm,Artificial Neural Networks(ANN)and Linear Regression algorithm.Here are the following conclusions:1)Compared with the other four existing snow depth retreiveal algorithms,the snow depth retreiveal algorithm proposed in this study performs best.The snow depth estimates have high accuracy,higher correlation coefficient,smallest MARE,MAE and RMSE.Among the three linear algorithms(Chang algorithm,SPD algorithm,and linear regression algorithm),the linear regression algorithm can generate higher precision snow depth estimates in deep snow.In general,the snow depth estimation capabilities of the nonlinear algorithms(ANN algorithm and SVR algorithm)are better than the linear algorithm.The snow depth retreival algorithm proposed in this study can decrease the impact of snow-saturation effect(usually the threshold of snowcovered saturation is 60 cm,and the threshold for snow-covered saturation in this study is increased to 150cm).2)The error in estimation of snow depth when the underlying surface is forest and shrub is relatively higher when the underlying surface is bare land and grassland.This is mainly because the vegetation attenuates the microwave scattering signal,resulting in underestimation of snow depth.In addition,when the snow is in stage ii(snow cover stabilization stage,from December to February),the snow depth retrieval model based on the four land cover types has a smaller MARE than the other two snow cover stages(snow cover accumulation stage,code i,from September to November;snow cover ablation stage,code iii,from March to June).In addition,the MAE and RMSE will change with time in each underlying situation(from snow cover stage i to stage iii).The snow depth retreival model based on the four land cover types will produce a smaller error in the snow cover stage iii.The proposed snow depth retrieval algorithm has been partially revised in this paper and applied to invert the daily snow depth data set from 1992-2016 in the Northern Hemisphere.Comparing the snow depth estimation accuracy between theproposed snow depth dataset and the existing two snow products(Glob Snow-2 and ERA-Interim/Land snow product);then we analyzed the characteristics of spatiotemporal snow cover changes in the Northern Hemisphere from 1992 to 2016 by using the snow depth data generated by this research and concluded that:1)Combining the daily snow depth measurement data from the meteorological stations in December and February,we compared the snow depth product generated by this researchs with the other two snow products(Glob Snow-2 and ERA-Interim/Land).After comparative analysis,it suggested that the accuracy of the estimated snow depth of Glob Snow-2 was the highest,followed by the SVR snow depth product(that is the snow depth product generated in this study),and finally ERA-Interim/Land snow depth product.2)We analyzed the variation of snow cover reserves over the past 25 years in the Northern Hemisphere.We found that annual snow cover storage showed a significant decrease in total snow reserves from 1992 to 2016,with 5500 km3/year.When analyzing the year-on-year changes in snow cover storage,it was found that the rate of decline in snow cover storage in January was the fastest,at 1065.72km3/year,followed by the snowfall reduction rate of 1060.10km3/year in November,and the slowest rate of decline in April with 128.04km3/year.In addition,we analyzed the interannual trends of snow cover storage in autumn,winter,spring,and summer,and found that the rate of snow cover storage interannual change in winter(from December to February)was significantly higher than that in the other seven months.They are 979.71 km3/year,1065.72 km3/year,and 738.79 km3/year,respectively.But the rate of change of snow cover storage in spring and summer is relatively minimal.3)The multi-year average snow cover duration shows strong latitudinal zonality and the shorter snow cover is mainly at the mid-latitude regions(25°N-45°N),such as China's East China,Central China,North China and Tarim Basin regions,the Mongolian Plateau,Western Europe,and most of the United States.The long snow cover duration is mainly located in the polar regions such as Alaska and northern Canada,northern Russia,and the Tibetan Plateau.4)From the distribution of the average snow depth in the Northern Hemisphere for many years,It can be seen that the snow depth in the Northern Hemisphere has obvious latitudinal zonality,and the snow depth will be deeper with the northward shift of latitude.For areas in China with many years of average snow depth greater than zero,they are mainly located in China's three major snow areas: the Qinghai-Tibet Plateau,northern Xinjiang,and northeast of China region.From the spatial distribution map of the average snow depth change rate in the Northern Hemisphere from 1992 to 2016,it can be seen that the snow depth program for most of the area in the Northern Hemisphere where the average snow depth is nearly half is increasing,and the rate of increase is 0-1cm.The range of snow depth decrease also accounts for nearly half of the northern hemisphere area,the rate of snow depth decrease is 0-1cm/year,and the fastest increase in snow depth decrease rate is greater than 1cm/year,and it is mainly located in the northern Tibetan Plateau in the central part of the western Tibetan Plateau.