Study On Spatio-temporal Variation Of PM_2.5 Concentration Based On Satellite And Ground-based Data Over Shanghai

In recent decades,environmental problems have become the focus of China and world,because serious-damaged ecological system and frequent-occurred air haze pollution.China's environmental bulletin showed that more than 70%cities which are involved in the monitoring exceeded the standard for pollutants,and 30%days per year failed to the standard during 2013 to 2017.Especially,PM_2.5?particulate matter,PM?was the main pollutant on severe-pollution days.Meantime,PM_2.5 can cause environmental changes,such as reduced visibility and imbalance of earth's radiation,and have a negative impact on human health.Therefore,the research on PM_2.5 has become the people's focus in recent years.China is limited for large-scale PM_2.5monitoring data before 2013 and ground monitoring stations.With the rapid development of remote sensing technology on multi-platform multi-sensor,multi-resolution,short revisit time and large space range observation,it can make up for the limits of ground monitoring.Shanghai is one of the cites for national air quality monitoring,because it is an important component of the Yangtze river delta region which is characteristic with rapid economic development and large population density.An aerosol optical depth?AOD?algorithm was firstly presented for visible remote sensing.Then,the model for PM_2.5 daily concentration was established by combined with stepwise regression and L-M optimized BP neural network method using AOD,meteorological,NDVI and time data.Further,the wavelet analysis method was used to mine the variation characteristics on different time scales for PM_2.5 concentrations in Shanghai.Finally,the spatial mappings of PM_2.5 concentration on different time scales were displayed and the reasons of spatial formation were explored with space correlation analysis.The main work of this paper is summarized as follows:?1?Based on the image dehazing processing,a retrieved algorithm for AOD was proposed.The algorithm based on the theories of dark channel algorithm and gaussian filter and can apply on visible remote-sensing images.The results showed that the algorithm can add the effective data and reflect the overall distribution of AOD in the study area.There is a correlation coefficient of 0.71 and a good inversion accuracy between retrieved AOD and ground measurement.Moreover,the spatial distribution of AOD in the study has the characteristics of gradually decreases from coastal areas to inland areas,high AOD?>0.9?in the city center and nearby Shanghai,and low AOD?<0.6?in water areas and surrounding areas.?2?A high-precision PM_2.5 concentration model was established by combining stepwise regression and L-M optimized BP neural network method.The performance of multiple linear estimation model for PM_2.5 concentration could be effectively improved by adding time factors,but its factors were exponentially increased.The stepwise linear regression method was used to select 23 important factors form 34factors.Then,L-M optimized BP neural network method was used to establish a high-precision PM_2.5 concentration model with the selected factors.The results showed that the combined model for PM_2.5 concentration has the good performance with a regression coefficient of more than 0.79 and a mean square deviation of 12?g/m³.Meantime,there is a regression coefficient of 0.76 between monthly PM_2.5concentration and ground measurement.The model showed high accuracy with regression coefficients from 0.81 to 0.88 at different sites.?3?During 2000 to 2016,the long-term characteristics of PM_2.5 concentration were analyzed in Shanghai.The results showed that air quality in Shanghai is good about79.92%,heavy-polluted about 0.38%of all days.There are three time-scale periodic characteristics for PM_2.5 concentration.The first main cycle is a stable cycle with the period length about one year,and PM_2.5 concentration shows a slight decreasing trend.The second cycle is a stable cycle with the period length about 6 years,and PM_2.5concentration shows a slight increase.The third main cycle shows complex changes with the period length about 6 months,and PM_2.5 concentration shows a decreasing trend.?4?The spatial distribution and variation characteristics of PM_2.5 concentration in Shanghai were analyzed.The mean of annual PM_2.5 concentration is mainly within the range of 48-56 g/m³ in different regions.There is a trend of spatial distribution that is gradually decreasing from coastal to inland and from east to west,which is mainly affected by the wind speed,direction and temperature in different regions.The variation of PM_2.5 concentration inter years is slightly different.Further,high PM_2.5concentrations in years were affected by a few factors,such as wind speed,wind direction,air temperature and boundary layer height.While low PM_2.5 concentration in years is mainly affected by the combined factors.The spatial distribution of monthly PM_2.5 concentration varies greatly from 27⁹2 g/m³,and the influencing factors are greatly different.The main innovations of this study can be summarized as follows:?1?A retrieved AOD method is provided based on image processing.The new algorithm overcame the limited of previous retrieved AOD algorithms about bands and low spatial resolution.?2?Considering delay effect of influence factors,factors of the two days were added to the original model.The results showed that 60%PM_2.5 concentration was affected by factors of previous two days.To some extent,the new method can improve the accuracy of PM_2.5 concentration.?3?There are three-time scales of PM2.5 concentration variations by wavelet transform method.The period lengths are 1 year,6 years and 6 months,respectively.The trend and characteristics of periodic variations were analyzed.