Research On The Inversion Method Of Atmospheric Boundary Layer Height Based On Lida

The atmospheric boundary layer is the region where energy,moisture,matter and momentum are exchanged between the surface and the free troposphere,and it is the main air layer constituted by human activities and each ecological environment.The boundary layer height is an important parameter to characterize the atmospheric boundary layer and an important physical parameter in the numerical weather prediction models and air quality assessment,and its accurate inversion is of great significance for meteorological science research,aviation and navigation,disaster mitigation and prevention.Lidar is widely used as one of the main detection methods for atmospheric boundary layer height.This dissertation addresses the problem of accuracy of lidar-based boundary layer height estimation from the perspective of signal processing,including the lidar-based boundary layer height inversion method and the signal pre-processing method before inversion,to provide valuable references and paths for further improving the accuracy of lidar-based boundary layer height inversion.The main work of this dissertation includes the following three aspects:(1)Research on lidar backscattered signal preprocessing method.The laser is inevitably disturbed by various noises in the transmitting and receiving process,and this interference is one of the factors affecting the accuracy of the boundary layer height inverted from lidar data.For the noise interference problem,a signal preprocessing method based on variational mode decomposition and Gaussian process is proposed.In this method,a new fitness function is designed to evaluate the denoising results based on variational mode decomposition.The relationship model between decomposition parameters and fitness function is established by Gaussian process,and the optimal parameters are selected.The reconstructed signal is further processed by using an interval threshold technique with global characteristics of the signal.The experimental results show that the proposed method outperforms other denoising methods based on empirical mode decomposition and variable mode decomposition,and can reduce the interference of noise on the signal.(2)Research on unsupervised learning method for retrieving atmospheric boundary layer height from lidar data.In order to overcome the influence of complex atmospheric conditions such as clouds and suspended aerosol layers on the inversion of boundary layer height,the K-mean atmospheric boundary layer height inversion method based on de-characteristic correlation and weight distance are proposed.Firstly,K-means initial parameters are selected based on the gradient characteristics of the lidar backscattering signal.Then the nearest distance is used to update the center to reduce the influence of singular values.Finally,the boundary layer height is retrieved based on the de-feature correlation and weight distance respectively.The experimental results show that the proposed method can better track the daily variation of the boundary layer height than the commonly used lidar boundary layer height inversion method,and the boundary layer height inverted by the proposed methods is in good agreement with that inverted by radiosonde.(3)Research on supervised learning method for retrieving atmospheric boundary layer height from lidar data.Unsupervised learning has better real-time performance,but the stability of learning effect is not strong.Supervised learning has a good effect and high stability,but it has high requirements for training data.To solve this problem,a joint inversion method of boundary layer height based on wavelet and random forest is proposed.The candidate values of boundary layer height are obtained based on the wavelet covariance transformation method,and the boundary layer height output of the random forest model is used to screen the candidate values,and the candidate value closest to the model output value is taken as the final boundary layer height.The experimental results show that the proposed method improves the reliability of the boundary layer height inverted by the wavelet covariance transformation method and reduces the sample requirements of the random forest model.The retrieved boundary layer height is basically consistent with the radiosonde measurement value.Through the above three aspects of research and experiment,it is proved that the signal preprocessing method based on variational mode decomposition and Gaussian process,the boundary layer height inversion based on de-feature correlation and weighting distance,the boundary layer height inversion based on wavelet and random forest,can improve the inversion accuracy of boundary layer height.The purpose of the study have achieved.