| Snow,as the most active element of the cryosphere,is an important indicator of the global climate change system.Snow water equivalent,expressed as the height of liquid water after snow melts,is an important input parameter in hydrological processes,ecological models,and climate systems,which plays an important role in the energy transfer process between soil and atmosphere.Therefore,obtaining the spatiotemporal distribution information of large-scale snow water equivalent is a work of great research value,a prerequisite for understanding global climate change,surface energy balance,and regional water cycle,and an urgent requirement for people in snow-covered areas to improve production and life safety.At the same time,it is also a major demand for the country to significantly improve water resources management,respond to global climate change,and improve disaster prevention capabilities.Passive microwave remote sensing has a physical basis for inversion of snow water equivalent due to its strong sensitivity to extinction processes such as scattering and absorption within the snow layer.In addition,passive microwave remote sensing has many advantages such as large observation scale,high temporal resolution,less affected by weather such as clouds and fog,and continuous observation all day and all weather.It is the most effective method for monitoring snow water equivalent for large-scale retrieval at present.The accuracy of the snow water equivalent inversion is mainly affected by the shading of the forest canopy and the temporal and spatial changes of the snow characteristic parameters.This paper starts from the parameter sensitivity of the snow radiation transfer model to explore the key snow cover characteristics that affect the snow water equivalent inversion.A dynamic snow water equivalent retrieval algorithm considering the change of snow particle size and a snow water equivalent inversion algorithm combining radiation transfer model and machine learning are developed.The main research conclusions of the paper mainly include the following four aspects:(1)Based on the EFAST global sensitivity analysis method,a detailed quantitative analysis was carried out on the change of the sensitivity of snowpack characteristic parameters in the snow radiation transfer model and the domain characteristics.Sensitivity analysis of snow model parameters is the premise of snow depth and snow water equivalent retrieval.The results show that,snow particle size,snow depth and snow density are the three main parameters that affect the microwave brightness temperature.Snow particle size and snow density are two important factors that affect the inversion of snow depth and snow water equivalent.When the snow particle size is less than 0.15 mm,the sensitivity of microwave radiation to snow depth is small,and the retrievability of snow depth is poor;when the snow particle size is between 0.15-0.3mm,the sensitivity of microwave radiation to snow depth is high,the retrievability of snow depth is strong;when the snow particle size is greater than 0.3 mm,the sensitivity of microwave radiation to snow depth decreases,and the retrievability of snow depth also decreases.In addition,due to the influence of snow pack metamorphism,the temporal sensitivity of snow pack parameters in the snow season changes dynamically,in which the snow density has a greater influence in the later period of the snow season,and the uncertainty of the snow density is another important source of retrieval error for snow water equivalent.(2)Three reanalysis snow density products(ERA-Interim,ERA5,and ERA5-Land)were verified based on the ground observation experimental data in China.The research results show that the snow density is very important for the deep snow in the late snow season,and the sensitivity of microwave radiation to the snow density accounts for more than 10%.The accuracy evaluation was carried out to find the product with better accuracy to replace the empirical value and use it in the snow depth and snow water equivalent inversion algorithm,which can improve the accuracy of the snow water equivalent inversion.It was found that the ERA-Interim snow density dataset has poor accuracy performance and significantly overestimates the snow density throughout the snow season;the ERA5 and ERA5-Land snow density datasets have similar performance and much better performance than ERA-Interim.The accuracy of the ERA5 snow density dataset is higher than that of the ERA5-Land dataset.The mean root square error RMSE of ERA5 dataset based on the snow course "point" observations is 56.2kg/m~3,and the RMSE based on the field dense network observations is 28.3kg/m~3.The ERA5 dataset can be used to replace the empirical values currently used for snow depth and snow water equivalent retrieval.(3)Aiming at the effect of forest canopy on microwave radiance temperature,a comprehensive radiative transfer model coupled with soil-snow-forest-atmosphere was established.The parameter of equivalent snow particle size,which characterize the microstructure of snowpacks,was retrieved based on the comprehensive radiative transfer model.First,considering the multi-elements of natural vertical structure,a comprehensive radiative transfer model including snow,forest,and atmosphere is developed.In the model,the atmospheric semi-empirical model was used to calculate the atmospheric transmittance of microwaves and the background radiation of the up and down atmosphere.According to the τ-ω model combined with ground observations and satellite observations,the optimal microwave transmittance of needle forests is calculated.Combined with the microwave transmittance of the existing broad-leaved forest area,the microwave transmittance of forest pixels was obtained by the method of mixed pixel decomposition.The air temperature of ERA5 at the surface of 2m was used to describe the emission information of forest upward.Then,a method for iterative optimization of the inversion of equivalent snow size by comprehensive radiative transfer model and satellite observation brightness temperature was proposed.(4)Aiming at the uncertainty of snow water equivalent inversion caused by changes in snow particle size and snow density,a model-driven dynamic snow water equivalent inversion algorithm and a snow water equivalent inversion algorithm combining radiation transfer model and machine learning were developed.Aiming at the problems of the existing snow water equivalent inversion algorithms that generally lack the evolution information of snow particle size,the algorithm has obvious underestimation for deep snow,and the retrieval error caused by assuming the snow density is a fixed value.The ERA5 snow density dataset predicted by the snow process model and the equivalent snow particle size obtained by iterative optimization inversion based on the integrated radiation transfer model were introduced into the paper.A dynamic snow water equivalent inversion algorithm considering the change of snow particle size and a snow water equivalent inversion algorithm combining radiation transfer model and machine learning were developed respectively.The research results show that the accuracy of the machine learning algorithm is not better than that of the model-driven dynamic snow water equivalent inversion algorithm without considering the snow parameters.Adding the equivalent snow particle size and snow density data can significantly improve the accuracy of machine learning inversion.The snow water equivalent inversion algorithm based on the combination of radiative transfer model and machine learning has high accuracy,which is obviously better than the current international remote sensing and reanalysis snow water equivalent products. |
PDF Full Text Request