Reserch On Neural Network Algorithm Retrieval For Ground-Based Atmospheric Paremeters

Atmospheric temperature profile, water vapor density profile and the cloud liquid water content are very significant to the analysis of climate change, weather forecast models and so on. On account of the ground-based microwave radiometer has high resolution on time, can be operated easily and conveniently, even if there is no one guarding it for a few months, it can still work continuously and has an accuracy detection, the ground-based mircrowave radiometer is now widely applied in the sensing of atmospheric section, the cloud liquid water, rainfall and so on[1-2]. As neural network algorithm has strong nonlinear mapping ability and parallel processing ability, it has been widely used in the atmospheric inversion[3].This paper studied on the algorithm of BP(error back Propagtion) neural network for ground-based microwave radiometer temperature profile, water vapor density profile and cloud liquid water content inversion, then implement and verify it by python programming language. Then concretely analyzed the ground-based microwave radiometer for atmospheric radiative transfer theory and the basic principle of atmospheric parameter inversion. Studied the principle of BP neural network algorithm and its improved algorithm for retrieval of atmospheric parameters, and using python language to implement and verify the momentum BP algorithm and LMBP algorithm. After that, the neural network algorithm and linear regression algorithm are applied to the atmospheric parameters inversion, than analysised and compared the performance: first comparing the two methods of temperature and humidity profile inversion results by using sunny data, the contrast results show that the neural network is more suitable for the temperature and huimidity profile inversion because of its stronger nonlinear ability and more accuracy than linear regression method. Secondly, when retrieving the content of liquid water, first should categorize the weather data and chose the cloudy data as liquid water content inversion. Two kinds of methods of inversion results show that the network inversion error is smaller.On the basis of this, the paper analyses the neural network inversion methods for further improving the atmospheric parameter inversion precision: analyzed the influence of the surface parameters and cloud information on the accuracy of temperature profile,humidity profile and liquid water content. The inversion results show that if considering the surface parameters, the accuracy can be improved, if the clouds information be added, the accuracy can be better than before especially the humidity profile and liquid water content inversion. By retrieving temperature and humidity profiles in different reasons, examines the influence of season on the parameter inversion accuracy and regularity, the results can be as a reference for its data quantity is less. Finally, compared the temperature and humidity results between inversions and RPG, then analyse the reasons of the difference and provide future improvement measures.