Research On Retrieval Algorithm Of Terahertz Ice Cloud Sounding

Ice clouds have obvious effects on the earth's energy cycle and water vapor cycle,so the study of the physical structure of ice cloud particles is of great significance to atmospheric radiative transfer models,climatic and meteorological cloud models and disaster weather forecasting.The size of ice crystal particles in ice clouds is mainly concentrated in 20-600 microns.Compared with microwave,it can only detect larger particles,and infrared band can only detect smaller particles,terahertz band has a unique advantage in ice cloud detection because its wavelength range is close to the size distribution of ice cloud particles.In the past two decades,foreign countries have carried out a series of studies on the terahertz radiation characteristics of ice clouds,developed a number of airborne terahertz ice cloud detectors,and completed a series of airborne detection tests.Although the research of terahertz ice cloud detection has just started in China,it has been considered to realize terahertz ice cloud detection load on the next generation polar orbiting meteorological satellites,so it is necessary to carry out prior research on ice cloud inversion algorithm.based on this,the key requirements such as ice cloud load detection channel setting,load detection mode and detection accuracy analysis are analyzed.At present,most of the inversion studies of terahertz ice clouds detection treat ice phase particles as the same kind of particle,but ice clouds contain not only ice particles,but also graupel particles.The ice phase particles are divided into ice and graupel particles in the mainstream RT radiation transmission model,the mainstream cloud model in the NWP numerical model and some satellite atmospheric remote sensing data products.Hence,the terahertz ice cloud detection inversion algorithm developed in this paper realizes the inversion of ice and graupel particles in ice clouds respectively.According to the radiation characteristics of graupel and ice particles,the graupel is first retrieved in 183 GHz band,and then the graupel is used as the prior value to retrieve ice in terahertz band.In this paper,two inversion algorithms,neural network and one-dimensional variation,are implemented,and the validity and precision of the algorithm are verified by simulation tests.To meet the above requirements,this paper solves the key technologies such as the construction of ice cloud atmosphere database,the calculation of terahertz ice cloud radiation transfer,and the inversion of ice cloud parameters based on neural network and one-dimensional variation.a full-link simulation system is realized from forward modeling of terahertz detection brightness temperature data of spaceborne ice cloud to ice cloud parameter inversion.Through the inversion simulation experiment,the key problems of terahertz ice cloud detection are studied,such as the selection of ice cloud detection frequency channel,the optimization of ice cloud inversion algorithm,and the detection accuracy analysis of total ice cloud,size and profile parameters.A unique technical feature of this paper is that ice cloud particles are divided into ice and graupel particles in detection and inversion,while the current terahertz ice cloud detection studies only study the parameters of ice particles.The simulation results show that the ice cloud physical parameters can be obtained from the observed brightness temperature data of single and multiple terahertz bands,and the detection accuracy of ice cloud with the combination of multiple frequency bands is higher than that of single frequency band,which can provide a reference basis for the selection of load frequency,The root mean square error of GWP realized by ice cloud inversion algorithm is 9.23g/m~2,the root mean square error of G?Zme is 24.12?m,the root mean square error of G?Zme is 513.62m,the root mean square error of IWP is 8.06 g/m~2,and the root mean square error of I?Zme is 300.48m,which can meet the application requirements of the model.Graupel particles and ice particles will affect the brightness temperature of terahertz band.After neural network inversion,using one-dimensional variational algorithm can improve the inversion accuracy of graupel and ice density profile.The relative error of graupel density profile is between 20%and 60%,and the relative error of ice density profile is between 20%and 50%.