Research On The Improvement Of Calibration And Atmospheric Parameter Retrieval Of Ground-based Microwave Radiometer

The ground-based microwave radiometer can continuously and unsupervisedly measure the atmospheric parameters such as the temperature and humidity from 0 to 10kilometers altitude,which makes up for the shortcomings of discontinuous observation time and high cost of the air sounding ball,and does great significance for improving the monitoring accuracy of small and medium-scale weather systems and the ability of strong weather forecasting.The measure performance of ground-based microwave radiometers depends on the accuracy and stability of atmospheric radiation brightness temperature measurement and atmospheric parameters retrieval,both of which are directly affected by the accuracy of calibration and neural network training.However,the current domestic ground-based microwave radiometer lacks in-depth research on the calibration error during the actual measurement process,and there is a lack of long-term observation brightness temperature training samples in the retrieval network training while using simulated brightness temperature for training is affected by the deviation from the observed brightness temperature.The above-mentioned problems jointly reduce the detection accuracy and restrict the localized deployment of microwave radiometers.To this end,combing the development of domestic ground-based microwave radiometers,this thesis analyzes and improves the factors that affect the calibration accuracy,and proposes an improved retrieval method based on data preprocessing to improve the quality and matching degree of observed brightness temperature and sounding data to achieve the purpose of improving the accuracy of inversion and actual observation ability.The main work done in this thesis is as follows:1.The theoretical research and correction of the injected noise deviation when the microwave radiometer observes different calibration sources are conducted,then the improvement effect of the deviation correction on the four-point calibration is verified through the liquid nitrogen and blackbody observation experiments.The result shows that the calibration accuracy is improved by 0.02?0.5K.Through experiments,the observation angles of the tipping curve calibration are optimized and improved.Using the observation angle combination of"low complementary elevation angle+high unilateral elevation angle"reduces the interference of atmospheric horizontal inhomogeneity while ensuring the reliability of calibration.After the calibration coefficient is improved,the error of the compared to the simulated brightness temperature on sunny days is reduced by 1.8?4.4K.2.A data preprocessing scheme for the observed brightness temperature is proposed,which includes sample classification,quality control and deviation correction,and is applied to the improvement of indirect sample retrieval.After data preprocessing,the correlation between the observed brightness temperature and the simulated brightness temperature has been significantly improved.Compared with the sounding data,the root mean square errors of the retrieved temperature,water vapor density,and relative humidity profile after preprocessing is reduced by 0.3?0.8K,0.06?1.09 g/m~3,3.9?4.3%,respectively,which effectively improves the accuracy of indirect sample retrieval and makes up for the lack of observation training set defects of direct sample retrieval,while still having considerable accuracy.The result shows that the indirect sample retrieval method based on data preprocessing is feasible in engineering applications.3.The co-site observation experiment with the foreign MP3000A radiometers verifies that the domestic microwave radiometer adopting data preprocessing to improve the retrieval of atmospheric parameters in the actual observation process is reliable,which has equivalent accuracy to the MP3000A.Based on the analysis of retrieval error,an improved data preprocessing scheme based on the classification of the training set is proposed.After this improvement,the retrieval accuracy of both sunny and cloudy samples is significantly improved compared to the retrieval results of unclassified training set and co-located observation experiments.