Utilizing Generalized Additive Models To Predict Air Pollutant Concentrations And Analyze The Effects Of Air Pollution Exposure On Daily Hospital Emergency Admission From Respiratory Diseases

In recent decades,China has experienced rapid growth in economic development,urbanization,and energy consumption.As a consequence,air pollution has become one of the major environmental problems in most megacities in China,and the health threats to urban residents due to air pollution are becoming increasingly severe.Many studies have revealed that long-term exposure to high concentrations of air pollution(especially PM_2.5)can significantly increase the risk of morbidity and mortality.Prediction of air quality is of great significance to the establishment of air quality assessment system and protection public health.The models usually used in prediction of PM_2.5concentrations,such as chemical transport model and neural network model,need to incorporate multiple types of prediction factors（such as air pollutants’source emission list,meteorological conditions,terrain data,etc.）,and determination of relevant parameters in the models are also complex.These constraints can limit the usefulness of these models.Therefore,it is important and necessary to develop simple,reliable and low-cost PM_2.5 prediction models when the data cannot be fully obtained or the data resources are limited.In this study,two cities with different urban topography and geographical location—Chengdu and Wuhan,were selected as the study areas.Using air pollutant data and meteorological data,trends and causes of air pollutant concentration,potential sources of pollutants between 2014 and 2019 were studied;a non-parametric regression model that can quantify the nonlinear relationship between response variables and covariates-generalized additive model（GAM）was used to identify the key meteorological factors that affect the PM_2.5 concentrations in Chengdu and Wuhan.Based on that,a model was developed to use time variables and the key meteorological parameters to predict PM_2.5 concentrations.Moreover,residual test and cross-validation were used to test the reliability of the model.In order to further explore the health effects of short-term exposure to air pollution,GAM models（single pollutant model and multi-pollutant model）were used to study the impact of air pollution on hospital emergency visits of respiratory outpatients under the condition of controlling time,meteorological factors,day of week effect and holiday effect.The main results are as follows.（1）The annual average mass concentration of major air pollutants in Chengdu and Wuhan continuously decreased,and the air quality gradually improved between2014 and 2019.In 2019,the annual average concentrations of NO₂,SO₂,PM_2.5,PM₁₀and CO in Chengdu were 40,7,47,68μg/m³ and 0.81 mg/m³,respectively,which are32%,61%,50%,42%,and 31%lower than those in 2014.The annual average concentrations of NO₂,SO₂,PM_2.5,PM₁₀ and CO in Wuhan in 2019 were 44,8,46,73μg/m³ and 1.00 mg/m³,respectively,which are 23%,68%,44%,35%and 19%lower than those in 2014.During the study period,unlike other air pollutants,O₃ in both Chengdu and Wuhan showed a trend of rising in a fluctuating way.In 2019,the average annual O₃ concentrations in Chengdu and Wuhan increased by 14%and 10%compared to that in 2014,respectively.In addition,although the particulate matter pollution has been effectively controlled in recent years,the annual average concentrations of PM_2.5and PM₁₀ in both Chengdu and Wuhan still exceeded the annual standards of NAAQS-2012 and WHO guidelines.Haze episodes in winter were still frequent in both cities.In 2019,there were 50 days and 40 days with the daily average PM_2.5 concentrations in Chengdu and Wuhan that still exceedsed the national ambient air quality standard（grade II:75μg/m³）,respectively,which indicates that further efforts need be made to reduce pollutant emissions in order prevent PM_2.5pollutions.（2）The air pollution in Chengdu and Wuhan is affected by both local emissions and long-distance transportation.The PM_2.5 potential source contribution function,concentration-weighted trajectory and backward trajectory cluster analysis show that the potential sources of PM_2.5 in Chengdu are mainly concentrated in Chengdu,and the south,southeast and east of the Sichuan Basin,and the external transport of other cities in Sichuan Province has a greater impact on the PM_2.5in Chengdu.The potential source areas of PM_2.5 in Wuhan are mainly concentrated in Wuhan,the northern Hunan,the northern Jiangxi and the southern Anhui.In addition to local contributions,the impact of PM_2.5 external transport cannot be ignored.It is difficult to effectively solve the problem of local air pollution only by emission control in a single city.It is necessary to consider the joint prevention and control of regional air pollution and formulate a comprehensive control policy focusing on multiple source categories at the local and regional levels.（3）GAM was used to identify the key meteorological factors that affect PM_2.5concentrations both in Chengdu and Wuhan.Results show that the key meteorological factors that affect the average daily PM_2.5 concentrations in Chengdu and Wuhan are the same,but the effect of each meteorological factor is different,and most of the relationship between meteorological factors and PM_2.5 concentrations are non-linear.For Chengdu,the meteorological factors that have the strongest correlation with the daily average PM_2.5 concentrations are temperature,wind speed,relative humidity,precipitation and sea level atmospheric pressure 5 days ago(SLP_5d),average mixed layer potential temperature,the direction and distance of the backward trajectory.For Wuhan,the meteorological factors that have the strongest correlation with the daily average PM_2.5 concentrations are temperature,relative air humidity,wind speed,wind direction,precipitation,sea level atmospheric pressure 4 days ago(SLP_4d),water vapor mixing ratio,and the direction and distance of the backward trajectory.According to the F and P values in the model fitting results,temperature is the most important meteorological factor affecting PM_2.5 in both Chengdu and Wuhan.Using time variables and key meteorological factors as explanatory variables,PM_2.5concentrations as the response variable,the fitting results of the GAM are:for Chengdu,the adjusted correlation coefficient（R²）is 0.735,the selected variables can explain the change of 74.8%PM_2.5 concentrations;for Wuhan,the model fitted R² is 0.701,and the selected variable can explain the change of 71.3%PM_2.5 concentrations.Model residual tests and cross-validation results show that GAM fitting results are reliable.This shows that the GAM can be used to predict the average daily PM_2.5 concentration of different types of cities.（4）The impact of sea level atmospheric pressure（SLP）on PM_2.5 concentrations has a certain lag effect.Different cities have varied lag days due to their geographical location,urban terrain,and the distance from the sea.Through comparison of different model cases,it is found that for Chengdu,the correlation coefficient between SLP_5dand PM_2.5 concentrations is significantly higher than the correlation coefficient between SLP and PM_2.5 concentrations on the same day,for Wuhan,the correlation coefficient between SLP_4d and PM_2.5 concentrations is significantly higher than the correlation coefficient between SLP and PM_2.5 concentrations on the same day.The fitting results of the GAM models also showed that SLP_5d/SLP_4d had a greater influence on the PM_2.5 concentrations.Compared with the SLP,the sea level atmospheric pressure a few days ago is a better predictor for predicting the average daily PM_2.5 concentrations.SLP can be used to build a GAM model to predict air pollutant concentrations a few days in advance.（5）The results of single pollution model analysis showed that PM_2.5,PM₁₀,NO₂,and CO had a significant impact on the daily emergency admission of respiratory disease,and there was a certain lag effect.The biggest effect of different pollutants on the number of emergency visits to respiratory system diseases occurred in different lag days.PM_2.5 had the strongest impact on daily emergency admission of respiratory on the same day,ER（excess risk）and 95%CI（confidence）were 0.78%（-0.16,1.72）.The strongest effect of PM₁₀ on the daily emergency admission of respiratory was on moving-average lag（0-1）d(ER=0.64%,95%CI:（-0.42,1.70）.The effect of NO₂,CO and O₃ on the daily emergency admission of respiratory reached the maximum on no-lag day.NO₂ also had a significant effect on the daily emergency admission of respiratory diseases on 1 d lag and all moving-average lag days.The effect of SO₂ on the daily emergency admission of respiratory diseases reached the maximum at on moving-average lag（0-3）d（ER=1.32%,95%CI:（0.25,2.39））.In the lag effect analysis of the multi pollutant model,the combined effect of multiple pollutants is lower than the sum of the combined effects of each pollutant alone,but close to the highest single pollutant effect.The influence of pollutants on the daily emergency admission of respiratory diseases was the largest on one lag day.For every 10μg/m³increase of air pollutant concentration,the daily emergency admission of respiratory diseases increased by 1.66%（95%CI:1.01,2.31）.The results of stratified analysis showed that the elderly（≥65 years old）are more sensitive than the young（<65 years old）;women were more sensitive to air pollution than men.Based on the above results,the main conclusions drawn are:（1）The sea level atmospheric pressure has a certain lag effect on the PM_2.5 concentrations in Chengdu and Wuhan.Therefore,the sea level atmospheric pressure can be used as a predictive indicator to predict PM_2.5 concentrations a few days in advance.（2）The GAM constructed using time and meteorology variables can predict the daily average PM_2.5concentration of different types of cities,and the model is simple,effective,reliable,and low-cost.（3）Ambient air pollutants has a significant impact on the number of hospital emergency department visits for respiratory diseases,and there is a certain lag effect.At different levels of gender,age and season,air pollution has different effects on the number of hospital emergency department visits for respiratory diseases.The study analyzed the temporal change of air pollutants trends,causes for changes and challenges in the future in Chengdu and Wuhan,further explored the key meteorological factors and their impact effects that affect the air pollutant concentrations in different cities,established a simple,reliable,low-cost air pollutant concentration statistics prediction model,and studied the relationship between air pollution and emergency visits of respiratory system diseases at multiple levels.The results of this study can provide theoretical reference for understanding the trends of urban pollutant change and meteorological causes,establishing air quality prediction models and warning systems,presenting scientific basis for single and regional air pollution prevention and control,and providing more detailed basis for the study of the relationship between short-tern air pollution exposure and respiratory diseases.This study also provides a more detailed basis and helps to the government to formulate accurate strategies for respiratory disease prevention and public health protection.