Research On Retrieval Of Temperature And Humidity Profile Based On Ground-based Microwave Radiometer Data

In this paper,the brightness temperature of a TP / WVP-3000 ground-based microwave radiometer is comprehensively controlled and the quality control effect is analyzed by using the sounding data of Daxing District,Beijing from 2013 to 2017,combined with the radiation transfer model Mono RTM.Then,a set of retrieval model is designed for each retrieval of temperature and humidity profiles by using Back-Propagation Neural Network algorithm,and the temperature and humidity profiles are retrieved based on the brightness temperature data after quality controlling,and compared with the ground-based microwave radiometer LV2 product.Then combined with BPNN and genetic algorithm to carry out the atmospheric temperature and humidity profile retrieval(GA-BPNN),and compared with the results using only BPNN(ALL-BPNN).The main conclusions are as follows:(1)The observation of the ground-based microwave radiometer is easily affected by non meteorological factors,so it is necessary to carry out quality control on its observation brightness temperature.The use of rain / non-rain inspection,extreme value inspection,time smoothness inspection and radiation transfer model inspection can play a certain quality control effect.After the whole set of quality control system,the availability of the data of the ground-based microwave radiometer is 76 %.After the quality control,the average deviation and root mean square error of the brightness temperature and the simulated brightness temperature are smaller,and the consistency is higher,especially the quality control effect of the channel near 30 GHz and 51 GHz is obvious,and the overall quality of the observation data is greatly improved.(2)After the quality control and deviation correction of the brightness temperature of microwave radiometer,the accuracy of the atmospheric temperature and humidity retrieval model based on BP neural network algorithm is improved compared with the LV2 product,especially the relative humidity result.Compared with the LV2 data of microwave radiometer,the accuracy of temperature detection is increased by about 6.77% and the accuracy of relative humidity detection is increased by about 20.11% in 0-10km?(3)Because the accuracy of BPNN model is greatly affected by the training samples,if the data distribution of the test samples is inconsistent with that of the train samples,the generalization performance of the model will be low.The accuracy of GA-BPNN inversion model based on the training samples optimized by genetic algorithm is further improved.Compared with the LV2 data of microwave radiometer,the accuracy of temperature detection is increased by about 10.21% and the accuracy of relative humidity detection is increased by about 23.75% in 0-10 km.