| In recent years,air pollution has become the focus of global attention.It is of great theoretical and practical significance for the prevention and control of air pollution to study the time-space distribution,dynamic changes,sources and transmission paths,and forecasting methods of pollutant emissions in Medium and small scale regions.The atmospheric assimilation inversion method has been proved to be one of the effective methods to improve the accuracy of simulation and forecast.However,the uneven distribution of ground-based observation data limits the temporal and spatial resolution and accuracy of assimilation inversion studies.At the same time,there are few assimilation studies of fine particles at regional scale(PM2.5),and even less assimilation inversion of air pollutants prior emission inventory（prior generalized flux）.For this reason,this thesis takes Xuzhou area as the research object.By constructing hourly scale 3km×3km resolution flux input data,combined with remote sensing data and ground-based observation data,it achieved efficient assimilation algorithm of the Proper Orthogonal Decomposition based Four-Dimensional Variational Data Assimilation（POD4DVar）and the online coupling of the atmospheric transport model WRF（Weather Research and Forecasting model）-CAMx（Comprehensive Air quality Model with Extensions）,constructed the Regional High Precision Assimilation Inversion Model Based on Joint Assimilation of Ground-Satellite Multi-Source Data and achieved bidirectional assimilation of PM2.5concentration and flux.The main work and achievements of this thesis are summarized as follows:1)Combined with the data of Multi-resolution Emission Inventory for China（MEIC）developed by Tsinghua University,and by using the Smoke model,Arc GIS and other software,the hourly-scale atmospheric transport model D01 layer flux input data was produced.Based on the measured environmental data in Xuzhou,the temporal and spatial distribution coefficients were determined,the local hourly scale3km×3km resolution D02 layer flux input data in Xuzhou was generated,and the quality of the local inventory was verified.By using this set of data as the initial flux data-driven assimilation model,the relative stable output results in the initial state of this model was ensured.2)Through the long-term comparative study of satellite-ground observation data,this thesis analyzed the long-term changes of PM2.5 in the study area.The results revealed that the emission data in the MEIC inventory,the monitoring data of PM2.5ground station and MODIS aerosol data all showed that the emission of PM2.5 in the central and eastern regions of China had obvious seasonal changes,and with a decreasing trend year by year.At the same time,the quality of PM2.5 concentration data retrieved by remote sensing was verified and analyzed.At the urban scale,the correlation coefficient between the daily average PM2.5 concentration of ground observation stations and the daily average PM2.5 concentration retrieved by remote sensing in ten cities was stable,about 0.8,and the root mean square error was less than 30μg/m3,which provided more effective observation data for assimilation experiments.3)Based on the efficient assimilation algorithm of POD4DVar,a regional high-resolution PM2.5 assimilation inversion model was constructed,a large number of tests were carried out to parameters such as the number of disturbance samples,the length of assimilation window,and the number of nested layers of the model,which contributed to the best assimilation scheme and improved the spatial and temporal resolution and accuracy of PM2.5 concentration inversion.The correlation coefficients of PM2.5 concentration in April,July,November and December 2018 were respectively increased from former 0.563,0.635,0.625 and 0.625 before assimilation to 0.677,0.668,0.733 and 0.718（after assimilation）.The root mean square error decreased from 0.0352 mg/m3,0.0103 mg/m3,0.0364mg/m3,0.0598mg/m3 to0.0274mg/m3,0.0097 mg/m3,0.0365mg/m3,0.0425mg/m3,respectively.The root mean square error and mean deviation in April and December decreased by 22.16%and 28.93%,respectively.Especially in April and December,the decrease of root mean square error and average deviation was more obvious,which decreased by22.16%and 28.93%,respectively.4)By introducing the PM2.5 concentration data obtained from MODIS aerosol optical thickness inversion into the assimilation model,the joint assimilation of ground-satellite was realized.This further improved the simulation accuracy--after adding remote sensing inversion data,the correlation coefficients in April,July and November increased from 0.677,0.668,0.733 to 0.710,0.752,0.754,respectively.The root mean square error decreased from 0.0274 mg/m3,0.0097 mg/m3,0.0365mg/m3（before）to 0.0254 mg/m3,0.0082 mg/m3,0.0336 mg/m3（after）,respectively.Due to the good quality of remote sensing data in July,the optimization results in July were more obvious.5)The model has the bidirectional output function of PM2.5 concentration and flux.By comparing the difference between the posterior flux and the prior flux of PM2.5 output from the assimilation inversion of the model,it was found that the posterior flux in April,November and December reduced by 2.61%,5.11%and7.73%,respectively,compared with the prior flux after the assimilation of the observed values,while the posterior flux in July increased by 3.85%.The output results can reveal that PM2.5 emissions in China are mainly concentrated in North China and developed metropolitan areas,and are significantly higher in heating season than in other seasons.The simulated surface PM2.5 concentration shows obvious seasonal variation--the simulated mean value in autumn and winter exceeds100μg/m3,while in summer,especially in July,the concentration is low,about 30μg/m3.Compared with the observed values at the observation stations,it can be found that the simulated concentration values in autumn and winter are still higher than the observed values after assimilation inversion,but lower than the observed values in summer.This indicates that the background flux data of the driving model still have large errors.There are 75 diagrams,18 charts and 264 references in this thesis. |
PDF Full Text Request