Study On The Method Of Aerosol Vertical Profile Based On Multi-data Fusion

Tropospheric atmospheric aerosols have a significant impact on climate,environment and human life.It is necessary to monitor atmospheric aerosols by multiple means.The spatial and temporal distribution of aerosols varies greatly in the troposphere.Lidar has high spatial and temporal resolution,so it becomes an important tool to detect the vertical profiles of aerosols.Lidar provides rich information on the vertical distribution of atmospheric aerosols.However,there are still some errors in measuring aerosols,which mainly come from the methods and noise.The inversion of Mie lidar data requires the advance determination of the aerosol extinction coefficient boundary value and extinction-backscatter ratio.The boundary value is usually estimated by the echo signals,and the aerosol extinction-backscatter ratio is often set as a fixed value,which will lead to a large inversion error.The Mie signals measured by polarization lidar can obtain the aerosol extinction-backscatter ratio profile based on the extinction-backscatter ratio model of depolarization data.The previous model set the depolarization ratio thresholds of dust and spherical particles as fixed values,and the inversion results of this model still have errors.The inversion of Raman scattering lidar data doesn’t require the predetermination of the aerosol extinction coefficient boundary value and extinction-backscatter ratio.Theoretically,Raman scattering lidar can obtain more accurate aerosol optical parameters,but Raman signals are weak and there’s more noise in them.In the denoising methods of Raman signals,the wavelet transform has been widely applied,but it needs complex parameter setting to obtain the optimal wavelet parameters.The workload is large,which is not conducive to the batch processing of Raman signals.This dissertation improved the inversion accuracy of aerosols through the fusion of different data and methods.To avoid complex wavelet parameter setting procedures,and retrieve the troposphere atmospheric aerosol optical parameters in batches.The wavelet transform combined with generalized regression neural network was used to smooth the Raman lidar echo data.The aerosol extinction coefficient boundary value for the inversion of Mie signals was determined by the Raman data.The boundary value was also estimated based on the assumed backscatter ratio and Mie signals.The corresponding inversion results were compared and analyzed.The more precise depolarization ratio thresholds of dust and spherical particles were determined by the optimized extinction-backscatter ratio model.The aerosol optical depth retrieved by the previous model was compared and analyzed.The results were as follows.(1)Referring to the aerosol extinction coefficient retrieved by the wavelet denoising,the root mean square error of the extinction coefficient retrieved by wavelet transform combined with generalized regression neural network was less than 11.9%.And the root mean square error of the extinction coefficient between 4 and 6 km was 3.3%.Compared with the estimated boundary value,the aerosol extinction coefficient boundary value determined by the Raman data can reduce the relative error of the extinction coefficient by 22.5% in the inversion of Mie data.(2)In the optimized model,the depolarization ratio thresholds of dust and spherical particles at 532 nm were calculated as 0.351 and 0.1,respectively.The depolarization ratio thresholds of dust and spherical particles at 532 nm were assumed to be 0.31 and 0.05 in the previous model,respectively.The inversion results obtained by the optimized model were closer to the measurement results of the sun photometer.Their correlation coefficient reached0.80,which is 0.08 higher than the one obtained before model optimization.The root mean square error of aerosol optical depth retrieved by the optimized model can be reduced by 39.3%.