基于深度神经网络估算近地面细颗粒物浓度的研究
首发时间:2020-12-31
摘要:由于我国地面环境监测站点建立时间晚,且空间分布稀疏,极大限制了我国对PM2.5进行长时间和区域连续性分布以及变化过程等研究。研究表明,利用卫星遥感相关数据与PM2.5之间的关系估算PM2.5浓度可以有效地突破PM2.5观测在时间和空间上的限制。本研究获取2016年至2019年全国国家控制监测站点的PM2.5浓度数据、多角度大气校正(MAIAC)算法反演的气溶胶光学厚度(AOD)产品、欧洲中心提供的气象再分析资料(ERA5)、归一化植被指数(NDVI)等相关数据,采用深度神经网络建立PM2.5时空模型,建模过程中分析了各个特性因子的线性相关性、NDVI和经纬度对模型的影响。训练、验证发现本研究构建的模型在川渝地区的估算结果和站点观测值相关性较好(验证结果R2 = 0.866),该模型被用于估算2010年至2019年中国川渝地区1 km空间分辨率的每日PM2.5浓度分布,并对PM2.5浓度在时间和空间上的变化进行分析。结果表明:PM2.5高浓度主要集中在四川盆地(四川东部和重庆西部)。除了2011年至2014年PM2.5逐年升高和2016年的略微上升,近10年我国川渝地区PM2.5浓度整体呈下降趋势。在2014年川渝地区PM2.5浓度到达峰值(年均30.57μg/m3),其中重庆市年均突破50μg/m3。因此,本研究通过深度学习方法估算长时间序列的颗粒物浓度,希望能够为川渝地区与PM2.5相关的流行病学研究和环境治理提供数据支撑。
For information in English, please click here
Based on Deep Neural Network Estimating the Concentration of Fine Particulate Matter in Near-ground
Abstract:Due to the late establishment and sparse spatial distribution of ground environmental monitoring stations in China, researches on the long-term and regional continuous distribution and change process of PM2.5 in China are greatly limited. Studies have shown that using the relationship between satellite remote sensing data and PM2.5 to estimate PM2.5 concentration can effectively break through the time and space limitations of PM2.5 observations. This study obtained the PM2.5 concentration data of national control monitoring sites, the aerosol optical depth (AOD) product retrieved by the Multi-Angle Implementation of Atmospheric Correction (MAIAC) algorithm, the meteorological reanalysis data (ERA5) provided by the European Center, normalized difference vegetation index (NDVI), and other related data from 2016 to 2019. A deep neural network is used to establish PM2.5 spatio-temporal model, and the linear correlation of each characteristic factor, the influence of NDVI and latitude and longitude on the model is analyzed during the modeling. Training and verification found that the estimated results of the model constructed in this study in the Sichuan-Chongqing area have a good correlation with the station observations (verification result R2 = 0.866). This model was used to estimate the daily distribution of PM2.5 concentration with a spatial resolution of 1 km in Sichuan and Chongqing region of China from 2010 to 2019, and the changes of PM2.5 concentration in time and space were analyzed. The results showed that the high concentrations of PM2.5 were mainly concentrated in the Sichuan Basin (eastern Sichuan and western Chongqing). Except for the annual increase in PM2.5 from 2011 to 2014 and the slight increase in 2016, the overall PM2.5 concentration in Sichuan and Chongqing has shown a downward trend in the past 10 years. In 2014, the PM2.5 concentration in Sichuan and Chongqing reached its peak (average annual 30.57μg/m3), of which the annual average in Chongqing exceeded 50μg/m3. Therefore, this study uses deep learning methods to estimate the concentration of particulate matter in a long time series, hoping to provide data support for epidemiological research and environmental governance related to PM2.5 in Sichuan and Chongqing.
Keywords: PM2.5 deep neural network long time series variation characteristics
基金:
引用
No.****
动态公开评议
共计0人参与
勘误表
基于深度神经网络估算近地面细颗粒物浓度的研究
评论
全部评论0/1000