利用卫星遥感数据估算上海市近地面空气PM_(2.5)浓度的探索

Application of satellite remote sensing to ground-level PM_(2.5) concentration estimation in Shanghai

[目的]探索利用卫星遥感数据建立符合上海市实际情况的近地面空气PM_(2.5)质量浓度的估算模型,并对模型进行检验和评价,为解决目前PM_(2.5)对人群健康慢性影响研究中存在历史数据缺乏和现实数据时空局限性提供有效的方法支撑。[方法]利用美国航天局(NASA)发布的中分辨率成像光谱仪(moderate-resolution imaging spectroradiometer,MODIS)L1B1KM卫星遥感数据,采用V5.2算法,推算气溶胶光学厚度(aerosol optical depth,AOD),并对其进行垂直和湿度校正,计算近地面干消光系数(Edry),引入日均温度、日均气压、日均风速等与气溶胶扩散密切相关的气象因子,建立不同季节估算PM_(2.5)质量浓度的多元线性回归模型,用实测值对模型进行验证,并与已有文献的建模方法进行比较。[结果]上海市春夏秋冬四季AOD均值分别为0.57、0.48、0.41、0.72,季节变化明显,冬春两季较高;空间分布上市区高于郊区。将AOD进行垂直校正和湿度校正,得到Edry,按照四季分别建立了由Edry估算PM_(2.5)质量浓度的多元线性回归模型,结果显示Edry与PM_(2.5)浓度四季均呈正相关,日均气压、日均风速与PM_(2.5)浓度均呈负相关,日均温度冬春季与PM_(2.5)浓度呈正相关,夏秋季与PM_(2.5)浓度呈负相关;按照现有文献方法,发现不引入气象因素模型的拟合精度和预测精度分别在50%左右,本研究建立的包含气象因素的模型四季的拟合精度和预测精度均较高,分别达到80.74%~90.83%和70.58%~77.79%,优于已有文献的建模方法。[结论]本研究基于卫星遥感数据构建的PM_(2.5)质量浓度推算模型拟合精度和预测精度都较高,经比较评价优于已有文献的建模方法,可以用于上海市近地面空气PM_(2.5)质量浓度的估算,以解决目前上海市PM_(2.5)质量浓度监测与人群慢性健康危害研究中存在的PM_(2.5)质量浓度历史数据缺乏和现实存在的监测站点少,分布不均等导致监测数据时空局限性的问题。

[Objective] To repair the deficiencies on missing historic data and real-time data of the PM2.5 with temporal-spatial limitation in the research of chronic population health impact by PM2.5 through building, testing, and evaluating a model to estimate ground-level PM2.5 of Shanghai using satellite remote sensing data. [ Methods ] We retrieved the moderate-resolution imaging spectroradiometer （MODIS） L1B1KM data to derive Edry （dry extinction coefficient） via aerosol optical depth （AOD） by V5.2 algorithm and adjusted by height and relative humidity. We used PM2.5 concentration as dependent variable, and Eddy, average temperature, average air pressure, as well as average wind speed as independent variables to build a multivariate linear regression model, followed by validation with real data and comparison with the model built by reported method. [ Results ] The average value of AOD in spring was 0.57, 0.48 in summer, 0.41 in autumn, and 0.72 in winter. There was an obvious seasonal tendency that the AOD average values were higher in spring and winter, and the spatial distribution showed higher values in urban area than in suburban area. We built four multivariate linear regression models for four seasons： The Edry was positively correlated with the PM2.5 concentrations in four seasons; Both average air pressure and average wind speed were negatively correlated with PM2.5 concentrations in four seasons; Average temperature in winter and spring was positively correlated with PM2.5 concentrations, but that in summer and autumn was negatively correlated. We compared our models with the model built by reported method without weather factors, and found the latter＇s fitting precision and prediction precision were both around 50%, while our models reached 80.74%-90.83% and 70.58%-77.79% respectively, better than the latter one. [ Conclusion ] The models built in this article based on satellite remote sensing data have higher fitting precision and prediction precision than the model built by reported method and could be used for estimating the ground-level PM2.5 concentration of Shanghai. Our results provide a solution to the research of PM2.5 monitoring and related chronic population health impact tackling the lack of historical PM2.5 concentration data and the restricted temporal-spatial data due to limited monitoring spots and poor coverage.