【作者】 肖璐；

【导师】 George Christakos； 吴嘉平；

【作者基本信息】 浙江大学 ， 物理海洋学， 2018， 硕士

【摘要】 随着我国人口持续增长和经济发展,环境污染问题日趋严重。其中,大气中的PM2.5细颗粒物已成为呼吸系统、心肺疾病和免疫系统等主要致病污染物之一。在我国,PM2.5污染引起了社会各界的关注。近些年来,我国意识到空气污染的严重性,开始建设PM2.5监测网,但是仍面I临缺乏准确的历史数据和有效的污染物预测模型这两大困境。本研究基于遥感气溶胶、气象、地形等多种数据,建立了一种新型统计模型,并预测了 2015年11月至2016年2月间我国沿海与内陆PM2.5时空分布特征。本文的研究区涵盖我国大陆大部分区域,包括整个沿海地区以及除新疆、西藏、内蒙古、青海和黑龙江以外的全部内陆省市区。研究区面积约为452万平方千米,占全国总面积的47.08%;区内人口约为12.72亿,占全国总人口数的93%。由于研究区人口量大而且高度密集,其空气质量状况对于我国环境污染治理以及社会经济健康效益评估具有重要意义。为提高模型精度,本研究融合了 PM2.5、NO2、CO地面监测数据、MODIS V5.2版本的气溶胶数据、数字高程、人口、气象、土地利用和道路等数据,应用贝叶斯最大熵(Bayesian Maximum Entropy,BME)理论和地理加权回归(Geographic Weighted Regression,GWR)方法,分析我国沿海与内陆的PM2.5时空特征,并获得P M2.5时空预测图。GWR建立了监测站点PM2.5浓度值与上述环境因子的局部空间关系,并预测30 X 30 km2空间分辨率的PM2.5浓度值;然后将预测的PM2.5数值作为软数据,监测站点PM2.5数值作为硬数据,放入BME中进行时间和空间上的协方差函数拟合,预测3X3 km2空间分辨率的PM2.5浓度值;最后将预测的PM2.5浓度数据与地面监测数据进行十折交叉验证,得到BME-GWR模型的预测结果。对比以前的研究结果,由于新方法的融合特性,BME-GWR方法对中国PM2.5浓度的时空预测有进一步的提高,且验证结果优于先前的大部分研究。GWR不能用在一个动态变化的时间上,综合BME和GWR方法后,新模型能够从时间和空间上同时预测PM2.5浓度变化,弥补了 GWR模型的缺陷。GWR模型2015年11月至2016年2月的拟合结果R2分别是0.78、0.83、0.80、0.63,气溶胶和压强两种环境自变量的回归系数在时间和空间上都存在明显的地域差异性。空间上,PM2.5污染严重区域呈现聚集的趋势,PM2.5质量浓度较高的地方主要分布在东北部、华北南部、华中东部及四川盆地东部等区域,长江东南和西北部地区的污染程度一直相对较小;时间上,冬季1月份、2月份的PM2.5污染程度相对较低,11、12月份PM2.5污染最为严重。BME-GWR的十折交叉验证结果(R2 = 0.883,RMSE= 11.39 μg/m3)表明模型精度较高。以沿海与内陆对比为例,时间上来看,地面监测点的PM2.5月平均质量浓度水平呈逐月上升然后下降趋势,且PM2.5月均值主要集中在35～100μg·m-3区间;空间上来看,严重污染的区域随着时间由沿海转移到内陆,由东北向西南转移,而且长江以南地区一直是污染较轻的区域。主要由于入冬以来供暖增加,以及不利的气象条件,地势等原因造成。总体而言,BME-GWR模型不论在预测维度和精确度上,都有较好的结果,也能为相关环境部门提供污染治理决策上的支持与帮助。今后,本研究的工作重点应放在收集更高分辨率的AOD数据和完善优化基于卫星遥感数据的预测模型,以便更高效地用于其他空气污染物的时空预测。更多还原

【Abstract】 With the rapid increment of population and development of economic in China,the environment is getting worse and worse.PM2.5 in ambient air has become a major pathogenic pollutant causing respiratory,cardiopulmonary and immune disease.,PM2.5 pollution has drawn the attention of society in China.In recent years,Chinese government was gradually aware of the seriousness of air pollution and started to build the PM2.5 monitoring network.However,China still faced the two major difficulties in PM2.5 studies,i.e.,lacking accurate historical data and effective predictive models.Therefore,based on a variety of data such as aerosol,meteorology and terrain,this study used a new geostatistical model to estimate the spatial and temporal distribution of PM2.5 in coastal and inland areas of China from November 2015 to February 2016.The study area covered in the China’s mainland,including the entire coastal and all inland areas except Xinjiang,Tibet,Inner Mongolia,Qinghai and Heilongjiang.The research covered an area of about 4.52 million km2,accounting for 47.08%of the total area in China.The population of the research area was 1.272 billion,accounting for 93%of the total population.Due to such the dense population distributed in the study area,the present study may have a great contribution for guiding the environmental pollution control and assessing economic health benefits in our country.To improve the accuracy of the PM2.5 estimation model,we integrated data from different sources for constructing the model,including the ground monitoring data of PM2.5,NO2 and CO,the aerosol optical depth data of MODIS V5.2,digital elevation,population,meteorological data,land use and traffic data.Especially,this study used Geographically Weighted Regression(GWR)combined with Bayesian Maximum Entropy(BME)theory to analyze the space-time distribution of PM2.5 between coastal and inland areas of China and generated the space-time mapping of PM2.5 concentrations.GWR established the local spatial relationship between the PM2.5 concentrations and the above considered factors,can be depicted by GWR model,and GWR model was used to predict the PM2.5 concentration with the spatial resolution of 30X30 km2;then,the predicted PM2.5 values were used as soft data and PM2.5 monitoring sites were used as hard data,respectively,we used hard data and soft data to fit the spatiotemporal covariance function in BME and predicted the PM2.5 concentration at spatial resolution of 3×3 km2.Finally,10-fold Cross-validation method was used to test the performance of the BME-GWR model by comparing the predicted PM2.5 concentration with the ground monitoring data.The cross validation results showed that the BME-GWR method provided a further improvement on the spatiotemporal prediction of PM2.5 concentration compared to previous studies,due to the characteristics of the new method,the cross validation results of BME-GWR technology were superior to most of previous research results.The pure GWR model can’t capture the PM2.5 variation with time,but the integrated BME-GWR method can take into account the spatiotemporal correlation and could simultaneously predict the temporal and spatial variations of PM2.5 concentration to make up for the deficiencies of the GWR model.The fitting results of pure GWR model showed that R2 were 0.78,0.83,0.80,0.63(November to February),and the regression coefficients of AOD and pressure,two environmental variables had significant geographical differences in both space and time.Spatially,PM2.5 showed a tendency to aggregate in heavily polluted areas,high PM2.5 concentrations were mainly distributed in the northeast,southern of North China,eastern Central China and eastern Sichuan Basin,whereas the pollution level was relatively light in southeast of Yangtze River and northwestern China;timely,the winter pollution was relatively low in January and February,but PM2.5 pollution was most serious in November and December.The 10-fold cross-validation of the BME-GWR(R2 = 0.883,RMSE = 11.39μg/m3)showed that the model accuracy was high.By comparing coastal and inland areas,timely,the monthly average concentration of PM2.5 showed an increasing-decreasing trend.The monthly average PM2.5 in the study area mainly concentrated in 35～100 μg·m-3;spatially,the area of severe pollution showed a tendency of shifting from the coast to the inland and from northeast to southwest,and the area south area of the Yangtze River was always a less polluted area.These were mainly due to the increments in heating since entering the winter as well as unfavorable weather conditions and topography.Overall,the BME-GWR model showed good results in terms of forecasting dimension and accuracy,and could also be used to support and help the relevant environmental departments on decision-making of pollution control.In the future,work should focus on collecting and extracting higher-resolution AOD data and developing better prediction models based on satellite remote sensing data for spatio-temporal prediction of other air pollutants efficiently.更多还原