| Fine particulate matter with an aerodynamic equivalent diameter of 2.5?m or less,known as PM2.5,has been a research hotspot due to its high pollution level and adverse health impacts.There are abundant researches indicating that PM2.5 is associated with cardiovascular and respiratory diseases and contributes to human's risk of death significantly.Exposure measures people's contact with the pollutant in a period.Thus,exposure assessment of PM2.5 can be used to study the association between pollution and diseases quantitively,and accordingly provide the statistical evidences and parameters on its health risk.Except for the risk of chronic exposure to PM2.5,researchers also found the effects of acute exposure,and suggested it's more accurate combining individual mobility with pollution to assess personal exposure dynamically.This method requires mapping PM2.5 concentration at a high spatiotemporal resolution.However,the sparsity and the uneven distribution of air monitoring sites pose a challenge.Against this background,we used satellite aerosol optical depth(AOD)and multiple random forest models to map full-coverage PM2.5 at a high spatiotemporal resolution,and then assessed the individual acute exposure in Beijing.This study is valuable to intraurban pollution control,suggestions on citizens'daily trips and subsequent epidemiological research as well as health risk assessment.The main content of this study is as follows:(1)Hourly Himawari-8 AOD(5 km)and daily MAIAC AOD(1 km)in Beijing from 2014 to 2019 were used in this study.Before they were added into the models,their quality and coverage were evaluated.Comparison with ground-based AOD showed that both satellite AODs had good quality.The coverage evaluation also indicated they were representative spatially and temporally.Thus,both satellite AODs can be used to estimate PM2.5.But AOD missing is significant and still an issue hindering full-coverage mapping.(2)We employed a random forest approach to estimate hourly PM2.5 concentrations at a 1-km resolution in Beijing from 2014 to 2019.Predictor variables included daily MAIAC AOD,hourly Himawari-8 AOD,hourly meteorological features,land use,etc.To utilize the spatial and temporal relations inside the data,we added the spatial convolution layers of PM2.5 observations and geographical features,and the temporal accumulation effects of meteorology to model inputs.To deal with the missing values in AOD and retain its information maximumly,we built 4 random forest models according to the availability of daily MAIAC AOD and hourly Himawari-8 AOD.Model validation showed fitting R2=0.975-0.982,RMSE=6.9-11.5?g/m3;10-fold CV R2=0.916-0.948,RMSE=14.1-16.9?g/m3;independent test R2=0.899-0.953,RMSE=18.7-19.9?g/m3.All four models had very good performances compared to previous research.Predictors'sensitivity test proved the contribution of all predictors in our study.Among the predictors,accumulation of meteorological terms which were seldom used in previous research turned out to improve the estimation of temporal trends significantly.(3)The priority of model estimates combination was determined according to the spatial trends estimation and resolution.The full-coverage hourly distributions of PM2.5during four pollution episodes in 2019 were mapped accordingly,demonstrating the details about pollution evolution and removal.The complexity of pollution's spatial and temporal characteristics affected by various mechanisms could be reflected and bring more dynamics to individual exposure assessment.(4)We combined individual mobility and hourly PM2.5 distributions to assess 313individuals'daily exposure and hourly peaks in 16 days of 2019.Ambient concentrations had been modified to be indoor or in-cabin concentrations by the ratios from other research.Individual mobility came from the open data shared by users of an app providing GPS services,and commuting trajectories were included.Then we analyzed the results of exposure assessment.Daily exposure and hourly peak exposure increased when the daily pollution level grew.Hourly peak exposure was often achieved when the hourly pollution was high or during the commute.For different individuals'daily exposure in one day,it's more correlated with the daily PM2.5 average at home than that at the workplace.While Beijing often showed the pollution pattern of“high south(-east),low north(-west)”,people who live in the northern and northwestern Beijing tended to suffer from lower daily exposure.As for the impacts from transportation modes,people who commuted by car often had lower daily exposure than those by walk or bike.Based on people's different commuting choices,we found that in order to lower the exposure,it's better to drive when the departure and the destination were of the same;when the mode was decided,it's better to reduce the commuting time.(5)We compared the differences between the static and the dynamic method to assess exposure based on data in Beijing.The static method refers to locating people at their home and ignoring the mobility.We found during the days with large spatial variation of pollution,people with long commutes(before concentration modification)and those with long commuting time(after concentration modification)often had nonnegligible biases when using the static method.Given the difficulty to collect individual mobility data,we proposed two simplified versions of the dynamic method:Method 1 requires the locations of home and workplaces(which can be collected through mobile phone data from telecom operators);Method 2 requires the activity schedules as well as the locations of home and workplaces(which can be obtained from activity diaries).When it's difficult to collect complete individual trajectories,the two methods can be applied to replace the dynamic method for the assessment of individual daily exposure,with higher accuracy than the static method. |
PDF Full Text Request