Study And Application Of The Retrieval And Numerical Optimization Methods Of Microwave Remote Sensing Atmospheric Profiles

The remote sensing technology has played a significant role in the meteorological fields for its advantages of high stability,reliability,intelligence and other characteristics.Meanwhile,the 24-hour continuous measurement of remote sensing technology has greatly enhanced the ability of meteorological monitoring which enables people to grasp the states of atmospheric movements accurately.However,the contact measurement can not be replaced with passive microwave remote sensing technology(PMRST)because of the low efficience and accuracy of PMSRT.High-precision PMRST has become one of the difficulties in the field of atmospheric detection.The topic of this dissertation mainly focuses on the atmospheric temperature and water vapor profiles in the range of 10 km above the ground.The measurement methods of atmospheric parameters and the development of PMRST are summarized in detail.In order to improve the accuracy and efficiency of PMRST,four key technologies including the physical regulars of microwave radiance transmission,the retrieval algorithms of atmospheric profiles,the algorithms of quality controlling of microwave sensing data and the relations between atmospheric profiles and microwave radiance are analysed.In the first part,the physical process of atmospheric microwave transfer is analysised.The dowelling microwave radiation transfer model(MRTM)is established and the variables affecting the transfer of microwave radiation are discussed based on the atomospheric physical regulars.The absorption coefficients and brightness temperatures(BTs)of different channels are calculated with the measured atmospheric profiles.On this basis,this part studies the characteristics of brightness temperatures at different frequencies and the atmospheric information of different altitudes to verify the feasibility of microwave remote sensing of atmospheric temperature and water vapor parameters.In addition,the microwave radiometer is introduced briefly as well as the experiment platform used in our study.In the second part,the retrieval algorithms based on neural network(NN)are studied in detail.The traditional algorithm regards the 10 km atmosphere as a whole,which results in low efficiency and accuracy of the NN.In order to solve this problem,traditional training algorithm is firstly derived.The results show that many outputs of NN converge with oscillations,which prolongs the convergence time greatly.The progressive training algorithm(PTA)is proposed to deal with this problem.Both of the cost function of NN and the outputs are taken into consideration in the PTA.According to the network parameters between two training iterations,the updated network parameters are directly determined by using local linearization and weighted mean methods.The PTA avoids the learning rate and eliminates the oscillations of network outputs effectively while training NN.In order to improve the efficiency further,the characteristics of 10 km atmosphere are discussed.The influences of clouds on the microwave radiance are taken into account.The error characteristics of the traditional algorithm are analyzed.Finally,the 10 km atmosphere is divided into three layers according to the error characteristics and the types of clouds.In accordance with the three layers,the layered retrieval algorithm is proposed.An experiment is designed to examine the feasibility and effectiveness of the proposed algorithms based on the radiosonde data.In the third part,we focus on the data quality controlling and numerical correction of the outputs of PMRST.The data quality controlling is a secondary examination of the atmospheric profiles of the PMRST to ensure that the output data conforms to the laws of atmospheric physics.It is found that the BTs calculated by the MRTM are not in good agreement with the BTs observed by microwave radiometer.The posterior numerical correction algorithm(PNCA)is proposed for this problem.According to the physical process of microwave radiation transmission in the atmosphere,the response of BTs of downwelling atmospheric microwave radiation to small change of atmospheric temperature at any altitude is analyzed.The weighting function between the BTs of downwelling atmospheric radiation and the atmospheric temperature parameters is derived and curves of weight functions of five channels in oxygen band are given.On the basis,the detailed mathematical model of the PNCA is established,and an experiment is designed with actual data.And,the performance of the proposed algorithm is evaluated with the results.The last part establishes the relationships between atmospheric conditions of rainy days and microwave radiation.The two-hour BTs before rainfall are studied to achieve the purpose of precipitation prediction.In view of the seasonal differences of BTs,the variation rate(VR)is proposed to achieve the consistency of BTs in different seasons.According to the difference of VRs between the clear and rainy days,four representative channels are selected for the nowcasting of rainfall events.The characteristics of two-hour VRs before rainfall are extracted by linear and exponential functions.In order to reduce the dimension and calculation of the prediction model,the correlation coefficients between the measured VRs and the characteristic equations are used to characterize the rainfall events.The NN and Bayesian network are adopted to nowcast the rainfall events.