Study On The Snow Cover Inversion Model And Its Application Based On The New Generation Advanced Satellite AMSR2 And VIIRS Data Fusion

Snow is the most important solid water resource in the study area-Xinjiang,and it is the main source of recharge in the Alpine river basin.Animal husbandry,industrial and agricultural use of water is highly dependent on the winter snow accumulation in the mountain,spring snow melt that forms the spring floods,which meet the urgent needs of the spring irrigation.Thus,it will provide unique water resources for local social and economic development and ecological environment evolution.Due to the importance of snow and ice,it is very necessary to study the spatial and temporal dynamic changes of snow parameters(snow cover,snow depth,snow volume,snow density)from the angle of research and application.Therefore,it is imperative to use the most advanced detection methods to accurately monitor the seasonal or permanent snow cover parameters in different watershed.Most of the conventional meteorological observation stations are located in flat towns or valleys,and their spatial distribution continuity is poor,which can't obtain the snow distribution information of high altitude areas in remote areas or in complex terrain.For the remote sensing monitoring of snow cover,the advantage of passive microwave frequency monitoring is that it can penetrate the weather phenomena such as clouds,fog and so on,can obtain the snow cover and snow depth quantitative information in real time.However,the low spatial resolution of the passive microwave data caused relative error by the rich heterogeneity within the pixel,hence the outline of the snow boundary is not clear,ean't monitor the light snow or patchy snow and so on.The advantages of visible and infrared monitoring of snow cover are the accurate identification of snow pixels and the distribution of snow cover with high spatial resolution.But it is not sensitive to the inversion of the thick snow and snow layers due to the weather conditions such as the cloud at day and night time.Hence,for considering the visible and infrared remote sensing snow simulations which can't retrieve the thick snow and are limited by atmospheric conditions,and passive microwave remote sensing is not sensitive to light snow and low resolution which caused the blurred edge of snow and bare land and other issues.In this article,we combined the visible/infrared and passive microwave data,that can be a complement of each other advantages of remote sensing information,and implement the retrieve of snow cover quantitative information within all-weather conditions,and proposed a new method,new idea and developed more reliable high-precision snow inversion model.In this present work,use the new generation of Earth observation satellite-AMSR2,NPPA/VIIRS remote sensing data and choose Xinjiang region as study area.The short infrared band of AMSR2 and VIIRS was fused by GS fusion algorithm,and the characteristics of the difference of brightness temperature and visible light infrared radiation of the snow cover,cloud,glacier,water body,vegetation,woodland,desert and wilderness etc.,and combine with the study areas terrain,geographical and seasonal characters,based on the recognition function and decision tree threshold method,the scattering index or polarization difference factor was constructed by collecting a large number of samples,and established the high precision snow inversion model with multi-source remote sensing snow parameters(Snow cover,snow depth,snow volume,snow density,snow surface temperature).The inversion accuracy of the new model was validated by using a large amount of snow data from all meteorological stations and field observations in the study area,and also compared the new model with the NASA algorithm.The results show that:(1)After the data fusion of visible infrared and passive microwave remote sensing data,the quality evaluation index shows that the brightness temperature of the new AMSR2 data image is sharper and the brightness gradient of the target information is increased,and the edge of the surface is becoming more clear,and could retrieve more underlying subtle surface targets.The spatial frequency of each band is significantly improved,the target structure in each band image has a clear and discernible contrast,and consistent with the high-resolution visible band image.Through the fusion process,such as snow,water,desert,bare land,terrain contours and other surface information are clearly visible,and the accuracy of the target parameters such as snow and snow depth etc.,are obviously improved.(2)Quantitative evaluation of snow cover and snow depth inversionThe snow coverage of the original AMSR2 inversion was higher-about 27.5%,the coverage of snow depth less than 10 cm was 11.3%?and after the fusion process it was 24.6%and ls.5%respectively.This model can be used to classify the information of snow and glacier,and estimate the thickness information of 1-60 cm depth snow cover,snow depth estimation are agree with the data from weatherstation and site observation with high accuracy.The complex correlation coefficient(R)was larger than 0.85,the root Mean Square Error(RMSE)was 2.92?6.9 cm and the mean absolute deviation was 2?4 cm.Accuracy for the 5 cm snow depth error was 91%?94%,for the 2.5 cm snow depth error was 81%to 87%.The RMSE of the quantitative evaluation index was relatively small,which indicates that the model is good for the time series inversion and suitable for snow inversion in different time periods.(3)The new model was compared with the NASA algorithm,there were relative difference between the two methods in different month,the accuracy of the new algorithm were higher than the NASA algorithm during the unstable snow accumulation period.The average error rate and the missed evaluation rate of the new algorithm were significantly lower than those of the NASA algorithm.The estimation of snow inversion in NASA algorithm was either too low or too high.Correlation coefficient of the new algorithm was relatively high and the average error was obviously low.(4)Accuracy evaluation of snow density inversion modelBased on AMSR2 microwave high-frequency polarization index or scattering index,the passive microwave snow density inversion model of the study area was established.The results show that the complex correlation coefficient R was 0.73,the RMSE was 1.418 g/cm3,and the average deviation Bias was 0.244 g/cm3.From the time series analysis,it can be seen that the simulation results were consistent with the field measurement results,especially in the winter season from December to next February,the simulated values were very close to the measured values.(5)Snow surface temperature inversion model and accuracy evaluationUsing the AMSR2 data,the model of snow surface temperature was established.The RMSE was 4.7 OC and 4.5 ?,and the mean absolute deviation index(MAD)was 3.7 ?and 3.319 ?,respectively.The correlation coefficient Rwas 0.88 and 0.91,respectively.(6)The total amount of snow in recent ten years decreased or close to the annual trend compared with the average value,especially in the winter and spring.The difference between high and low snow years was 400 billion cubic meters in winter,320 billion cubic meters in spring.The month of few snow accumulation accounts for 37%of total amount and the month of high snow accumulation accounts for 28%,respectively.The spatial distribution of snow covers in different season at different altitude are varied greatly.In summer and spring/autUmn,the area covered by snow are ranging from 5?8%and 17?22%,respectively,between the elevation of 3000 and 4500 m.Above the elevation of 4500m,snow area accounted for 45?75%and 56?81%of the total snow covered area.Winter snow cover is mainly distributed in Northern Xinjiang(full covered),Eastern Xinjiang and Southern Xinjiang(some parts of mountain areas are covered),the percentage of total snow cover area is about 53.5?70%.Winter snow depth less than 20 cm is mainly distributed in the area of lower than 1500 m elevation,and larger than 40 cm is mainly distributed in the middle and high mountains of 1500?3000 m elevation.The glaciers in the study area are mainly distributed in the high elevation areas of Hotan,Kashi,Kezhou,Aksu,Bazhou and Ili.The glacier area is about 1208?8766km2.In summary,the high-precision snow inversion model based on multi-source remote sensing infornation fusion has modified and improved the current snow parameter of inversion algorithm and product generation method,and established a high precision snow inversion parameter dataset,for regional ecosystem protection,water resources development and utilization and flood control and drought-resistant decision-making,providing scientific basis and practical application.