Special Session on Better Air Quality in Asia (I)

Yingying Zeng1, Daniel A. Jaffe2,3, Xue Qiao4,5, Yucong Miao6, Ya Tang 1,5

 

Department of Environment, College of Architecture and Environment & Healthy Food Evaluation Research Center, Sichuan University, Sichuan 610065, China
Department of Atmospheric Sciences, University of Washington, Seattle, Washington 98195, USA
School of Science, Technology, Engineering, and Mathematics, University of Washington, Bothell, WA, USA
Institute of New Energy and Low-carbon Technology, Sichuan University, Chengdu 610065, China
State Key Laboratory of Hydraulics and Mountain River Engineering, Sichuan University, Chengdu 610065, China
State Key Laboratory of Severe Weather & Key Laboratory of Atmospheric Chemistry of CMA, Chinese Academy of Meteorological Sciences, Beijing 100081, China


 

Received: November 14, 2019
Revised: February 12, 2020
Accepted: April 1, 2020
Download Citation: ||https://doi.org/10.4209/aaqr.2019.11.0586  


Cite this article:

Zeng, Y., Jaffe, D.A., Qiao, X., Miao, Y. and Tang, Y. (2020). Prediction of Potential High PM2.5 Concentrations in Chengdu, China. Aerosol Air Qual. Res. 20:956-965. https://doi.org/10.4209/aaqr.2019.11.0586


HIGHLIGHTS

  • GAMs were used to explore the relationship between meteorology and PM2.5 in Chengdu.
  • SLP five days earlier is a good predictor for PM2.5 in Chengdu.
  • A GAM prediction tool was developed with an R2 of 0.73 for daily PM2.5.
 

ABSTRACT


Daily exposure to high ambient PM2.5 increases the mortality rate and contributes significantly to the burden of disease. In basin-situated cities with high local emissions of air pollutants, meteorological conditions play a crucial role in forming air pollution. One such city is Chengdu, which is located in the Sichuan Basin and serves as the economic, educational, and transportation hub of western China. Particulate matter with an aerodynamic diameter of < 2.5 µm (PM2.5) is the most critical pollutant in this city. Although the annually averaged PM2.5 concentrations declined from 92 to 57 µg m–3 between 2013 and 2017, the city still suffers from haze and smog, with 85 days during 2017 displaying 24-h PM2.5 concentrations > 75 µg m–3. To better understand the influence of meteorological factors on PM2.5 pollution with the goal of easily and reliably predicting the latter, we examined the relationships between the 24-h concentration and a variety of meteorological parameters in Chengdu. We found that the strongest predictors of the PM2.5 concentration were the temperature, precipitation, wind speed, and trajectory direction and distance. Furthermore, although the same-day sea-level pressure (SLP) was a weak predictor, the SLP 5 days in advance performed better. We developed generalized additive models (GAMs) that predicted the PM2.5 concentration as a function of multiple meteorological parameters. One of the GAMs developed in this study exhibited an adjusted correlation coefficient (R2) of 0.73 and captured up to 73.9% of the variance in the daily averaged PM2.5 concentrations. The model performance was improved by using the ΔSLP (i.e., mean pressure difference) for 5 days instead of the SLP, suggesting that ΔSLP5d is a good predictor of high concentration days in Chengdu. This study provides a useful tool for controlling emissions in advance to prevent heavy pollution days and issuing outdoor activity warnings to protect public health.


Keywords: PM2.5; Meteorology; Air pollution; Generalized additive models; Air pollution prediction


INTRODUCTION


With rapid economic growth and concomitant increase in anthropogenic emissions during the recent decades, PM2.5 (i.e., particulate matter (PM) with an aerodynamic diameter of < 2.5 µm) pollution has become a great environmental and health challenge in China (Miao et al., 2019; Zeng et al., 2019). PM2.5 not only harms human health but also influences climate (Di et al., 2009; Mahowald, 2011; Tao et al., 2014; Bo et al., 2015). In 2017, nearly 3 million deaths worldwide were attributed to PM2.5 exposure (HEI, 2019). China and India lead the world in the disease burden attributable to PM2.5 (Lelieveld et al., 2015; Liu et al., 2016; Cohen et al., 2017; Liu et al., 2017). Short-term exposure to high PM2.5 concentrations is associated with specific health effects, such as respiratory and cardiovascular diseases, and increased mortality (Hoek et al., 2013; Chen et al., 2018). Children, females, the elderly (> 65 years old), and those people with pre-existing health conditions are most sensitive to PM2.5 pollution (Luo et al., 2018; Lei et al., 2019). In general, the effects of PM2.5 on daily respiratory and non-accidental deaths are more evident in the cold season than other seasons in China (Li et al., 2019; Singh et al., 2019).

In addition to emissions, topography and meteorology also play a significant role in temporal variations in regional air quality (Boznar et al., 1993; Palau et al., 2005; Saide et al., 2011; Hu et al., 2014; Gustin et al., 2015; Miao et al., 2015). Basin landform facilitates the accumulation of air pollutants to high concentrations (Miao et al., 2019). As the most populous basin in China, with a population of over 83 million in 2018 (Sichuan Statistics Bureau, 2018), the Sichuan Basin (SCB) has been identified as one of four most air-polluted regions in China, along with the Beijing-Tianjin-Hebei region, the Yangtze River Delta, and the Pearl River Delta (Lin et al., 2012; Tian et al., 2017). As one of the two megacities in the basin, Chengdu is located in the western boundary of the SCB (Fig. 1), with a population of over 16 million, over 4 million motor vehicles, and intensive industrial and agricultural activities, leading to high anthropogenic emissions (Qiao et al., 2019). To the west and east of Chengdu are the Longmen-Qionglai Mountains and the Longquan Mountains, respectively. Climate in Chengdu is subtropical humid monsoon climate, with a rainy season from June to August. Previous studies have investigated the impacts of meteorological parameters on air pollution, including ambient air temperature, air pressure, precipitation, winds, and boundary layer structure (Li et al., 2015; Zeng and Zhang, 2017; Miao et al., 2018). Many studies have revealed that PM2.5 pollution is highest in winter in Chengdu due to unfavorable meteorological conditions for air pollutant to disperse and deposit, such as shallow mixing layer, high relative humidity, and strong thermal inversion (Tao et al., 2013; Miao et al., 2018, 2019). Although the critical role of meteorology in PM pollution in Chengdu has been recognized, a comprehensive analysis of the overall relationship between meteorology and PM2.5 concentration has not been conducted.

Fig. 1. (a) Topographic map of Sichuan Basin and the location of Chengdu, and (b) the air quality monitoring sites in Chengdu. Site names are given in SI Table S1.Fig. 1. (a) Topographic map of Sichuan Basin and the location of Chengdu, and (b) the air quality monitoring sites in Chengdu. Site names are given in SI Table S1.

There are three major types of methods to forecast air quality, including statistical models (e.g., multiple linear regression (Ul-Saufie et al., 2012; Ng and Awang, 2018), artificial neural networks (Hooyberghs et al., 2005; Grivas and Chaloulakou, 2006; Wang et al., 2014), support vector machines (Lu and Wang, 2005; Wang et al., 2014), hidden Markov models (Sun et al., 2013a, b), grey system model (Lee et al., 2007; Pai et al., 2013), chemical transport models (CTMs; e.g., WRF-Chem and Community Multiscale Air Quality [CMAQ]) (Ryu et al., 2018; Wilkins et al., 2018; Goldberg et al., 2019; Yang et al., 2019), and a combination of statistical models and CTMs (Carmichael et al., 2008; Wang et al., 2016; Xue et al., 2019). Compared with statistical models, CTMs are much more complex and time-consuming and depend on high computing capability and a lot of input data, such as emissions inventories, topography, land use and cover, and meteorological simulations. The performance of CTMs is greatly influenced by the aforementioned input data and the chemical mechanisms used (Zhang et al., 2012). Statistical models directly fit the patterns derived from the observed data, which are well suited for quantifying the nature of pollutant response to meteorological parameters, and are simpler and might be more accurate compared with chemical transport air quality models (Schlink et al., 2006; Sun et al., 2013; Zhang et al., 2015). The generalized additive model (GAM) is a combination of a generalized linear model and an additive model (Hastie and Tibshirani, 1990) and has been widely used to explore the impacts of meteorological parameters on air quality (Aldrin and Haff, 2005; Asimakopoulos et al., 2012; Tian et al., 2014; Gong et al., 2017). The GAM can be used to understand the mechanisms that cause atmospheric pollution and to forecast air pollutants’ concentrations (Schlink et al., 2006; Hübnerová and Michálek, 2014).

In this study, a GAM was used to understand the relationships between 24-h PM2.5 concentrations and critical meteorological factors for the five years from January 2013 to December 2017 in Chengdu. It is important to understand the mechanisms of air pollution formation before we can provide recommendations for government to effectively control emission and to release early warnings for the public. The main objectives of this study are: (1) to find the key meteorological factors affecting PM2.5 concentrations by using GAM and (2) to investigate whether the GAMs built with meteorological data can be used to predict 24-h PM2.5 concentrations.


MATERIALS AND METHODS



Data Sources and Variables

Hourly PM2.5 concentrations from January 2013 to December 2017 were obtained from seven national air quality monitoring stations in the urban areas of Chengdu (via the China National Environmental Monitoring Centre: http://www.cnemc.cn/) and one monitoring site at the United States General Consulate in Chengdu. For data quality control, we used an automatic outlier detection method, which was based on the probability of residuals to exclude outliers due to power outages, instrument calibration, and/or other reasons (Wu et al., 2018). The results of the Pearson correlation analysis show that hourly PM2.5 concentrations at the eight sites were significantly related, with R2 values ranging from 0.892 to 0.971 and slopes close to 1 (Table S1). Therefore, the hourly PM2.5 concentrations measured at the eight stations were averaged to calculate citywide daily PM2.5 concentrations.

Standard hourly surface meteorological data (including air temperature, relative humidity, precipitation, sea-level pressure, and wind speed and direction) measured at the national meteorological station (Wenjiang Station: 30°44′N, 103°51′E, 547.7 m above sea level [a.s.l.]) were obtained from the China Meteorological Administration. Radiosonde meteorological data for Wenjiang Station at 8 a.m. and 8 p.m. local time (LT) were obtained from the website of the University of Wyoming, the United States of America (http://weather.uwyo.edu/upperair/sounding.html). The Hybrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT) model (version 4.9) was run for each day using the Global Data Assimilation System (GDAS) data (1° × 1°) to calculate 12-h and 24-h backward trajectories, respectively. The trajectories were initialized for times of 10 a.m. and 10 p.m. each day. Detailed descriptions of all the meteorological variables used in this study are provided in Table S2.


GAM Method and Model Development

GAMs are an expanded form of the generalized linear model (GLM). Instead of linear coefficients, GAMs involve a combination of flexible smooth spline functions of the covariates. The advantage of GAMs is that they do not need pre-set parameters and they can automatically select appropriate polynomial fits. GAM techniques have been previously applied to explore complex non-linear correlations with air pollution (Dominici et al., 2002; Camalier et al., 2007; Pearce et al., 2011; Bertaccini et al., 2012; Gong et al., 2017; Gong et al., 2018). We used the package mgcv in the R environment to develop a GAM as implemented by the gam() function (Wood, 2006):

where yi is the ith concentration of a given air pollutant, β0 is the overall mean of the response, sj(xij) is the smooth function of ith value of covariate j, n is the total number of covariates, and εi is the ith residual.

The smoothness of each function in the equation is controlled by the number of knots or the effective number of degrees of freedom. As the number of knots increases, the observed PM2.5 concentrations fit better. However, it will become less smooth and effectively over-fit the data. The default method of cross-validation was used to choose the degree of smoothing (penalization) of k = 10 dimensional regression splines (Wood, 2006). In order to select the key parameters out of a large number of parameters, the Akaike Information Criteria (AIC) was evaluated for each model specification. A variable remained in the final model if the fit yielded a lower AIC (Wood, 2006; Pearce et al., 2011). P-values could approximate confidence for inclusion in GAM, and F-values could be used to identify the strongest meteorological predictors. We chose year, month, and trajectory direction and distance as the basic parameters for the model, then added the other variables one by one to obtain the models with a lower AIC. The trajectory direction and distances on a single day were calculated from 12-h or 24-h LT between the starting and end point. We used log-transformed PM2.5 concentrations in the model to reflect the log-normal distribution of PM2.5 data. The model used for 24-h PM2.5 in Chengdu can be described as:

where Year and Mon are year and month, respectively. TEM, TMAX, and TMIN are daily average, maximum, and minimum of air temperature, respectively. RHMIN is daily minimum relative humidity. WS and WD are 24-h wind speed and wind direction. PRC is daily total precipitation. SLP5d is SLP 5 days earlier. WD8 is wind direction in lowest 1000 m height at 8 a.m. T1Km8 is air temperature in the lowest 1000 m at 8 a.m. RH1Km8 is relative humidity in the lowest 1000 m height at 8 a.m. MLTheta8 is mean mixed layer potential temperature in K at 8 a.m. MLMR20 is mean mixed layer mixing ratio in g kg–1 at 8 p.m. Tradist10a and Tradist22a are the distances between end points (point to point) after 12 hours of transport for a back trajectory initialized at 10 a.m. and 10 p.m., respectively.

The R gam.check function was used to assess the GAM’s performance. Specifically, the Q-Q plots (sample quantiles against predicted quantiles), scatter plots (residuals against linear predictor), histogram of the residuals, and scatter plots (response against fitted values) were generated. In addition, we randomly selected 10% of the datasets as a cross-validation dataset and 90% of the data as a training set to build the GAM model, and did model prediction and evaluated model performance.


RESULTS AND DISCUSSION



Characteristics of PM2.5 Pollution


Inter-annual Changes

The PM2.5 concentrations and the air quality levels (AQLs) based only on PM2.5 concentrations in each year from 2013 to 2017 are summarized in Table 1. The annual average PM2.5 concentrations declined from 92 to 57 µg m3 from 2013 to 2017. PM2.5 pollution was the heaviest in 2013, with 24-h PM2.5 concentrations > 250 µg m3 (AQL: Grade VI) and > 150–250 µg m3 (Grade V) on 7 and 49 days, respectively. The number of days with PM2.5 concentrations < 75 µg m3 (Grade I and Grade II) increased between 2013 and 2017, but heavy PM2.5 pollution days (24-h PM2.5 > 115 µg m3) gradually decreased. The decreased annual average PM2.5 concentrations reflect the great achievement in air pollution mitigation through relevant policies, regulations, and technologies (Jin et al., 2016; Zeng et al., 2019). However, 24-h PM2.5 concentrations still exceeded the national ambient air quality standard (NAAQS) (Grade II: 75 µg m3) on 85 days in 2017, thus more efforts should be made to further prevent PM2.5 pollution in Chengdu. 


Seasonal and Diurnal Variations

The diurnal patterns of PM2.5 concentrations during 2013–2017 in each season are presented in Fig. 2. The hourly averages of PM2.5 concentrations in the four seasons were in the order: winter (99–126 µg m–3) > spring (54–78 µg m–3) > autumn (50–66 µg m–3) > summer (37–50 µg m–3). The diurnal patterns were similar in the four seasons: PM2.5 concentrations peaked at 10–11 a.m., about 2–3 hours after the morning traffic rush hour. Then, PM2.5 concentrations decreased and reached a daily minimum at 5–6 p.m. The diurnal patterns probably reflect the impacts from inversion layer and air pollutant emissions. From midnight to 9–11 a.m., the city is affected by an inversion layer, which inhibits air pollutant dispersion. Thus, PM2.5 concentrations increase slightly during this period. Also, the morning traffic rush contributes to PM2.5 concentrations. As the air temperature increases, the inversion layer disappears at 10–11 a.m., thus PM2.5 concentrations start to decrease until 5–6 p.m. In parallel with the formation of a new inversion layer in the evening, air pollutants from the afternoon traffic rush around 6 p.m. and other sources cause PM2.5 concentration to increase from 6 p.m. (Tang et al., 2015; Tang et al., 2016). 

Fig. 2. Diurnal variation (local time) in PM2.5 concentrations in spring (March–May), summer (June–August), autumn (September–November), and winter (December–February) in Chengdu from 2013 to 2017.Fig. 2. Diurnal variation (local time) in PM2.5 concentrations in spring (March–May), summer (June–August), autumn (September–November), and winter (December–February) in Chengdu from 2013 to 2017.


Correlation of 24-h PM2.5 with Meteorological Factors

Fig. S1 presents the correlation of 24-h PM2.5 concentrations with the effect of each meteorological factor. We found that all the selected meteorological factors have a non-linear relationship with PM2.5 concentrations. Among the 24 meteorological parameters, air temperature, wind speed, relative humidity, precipitation, and ΔSLP (see definition below), mean mixed layer potential temperature, and trajectory direction and distance are good predictors for PM2.5 in Chengdu, with F-values between 3.533 and 210.939, and P-values < 0.01 in the model fit. PM2.5 concentrations were positively correlated with daily maximum temperature and the radiosonde air temperature in the bottom layer (< 1,000 m) at 8 a.m. PM2.5 concentrations were negatively correlated with air temperature, daily minimum relative humidity, wind speed, and precipitation. High wind speed and high daily average temperature favor the dispersion of air pollutants, and high relative humidity and precipitation facilitate PM2.5 deposition (Fang et al., 2017; Zhao et al., 2018; Lin et al., 2019). 

Fig. 3. Scatter plots of PM2.5 concentrations and sea-level pressure with quadratic fits: (a) PM2.5 concentrations vs. sea-level pressure on the same day; (b) PM2.5 concentrations vs. sea-level pressure 5 days in advance.Fig. 3. Scatter plots of PM2.5 concentrations and sea-level pressure with quadratic fits: (a) PM2.5 concentrations vs. sea-level pressure on the same day; (b) PM2.5 concentrations vs. sea-level pressure 5 days in advance.

We performed a correlation analysis on the relationship between sea-level pressure and PM2.5 concentration (Fig. 3). The results showed that the correlation coefficient between PM2.5 and the sea-level pressure on the same day is 0.1237. In contrast, the correlation coefficient between PM2.5 and the sea-level pressure 5 days earlier is 0.3603. To further explore the relationship between sea-level pressure and PM2.5, we selected the 30 highest PM2.5 concentration days in each year (total 150 days in five years), and then calculated the mean sea-level pressure difference (∆SLP: ± 95% confidence). We define ΔSLP as follows:

We conducted a statistical analysis on the relationship between the PM2.5 concentrations and sea-level pressures for 20 days (10 days in advance and 10 days’ lag, respectively) on the highest PM2.5 concentration days (Fig. 4). Mean ΔSLP 2–10 days in advance were always significantly different from zero, furthermore, the difference was most significant at a lag of 5–7 days, but mean ΔSLP for 1–10 days’ lag were not significantly different from zero. In other words, sea-level pressure 2–10 days in advance is a predictor for a high PM2.5 day in Chengdu, and the strongest predictive power was 5–7 days in advance.

Fig. 4. Mean ΔSLP (± 95% CI) on the 30 highest PM2.5 concentration days in each year.Fig. 4. Mean ΔSLP (± 95% CI) on the 30 highest PM2.5 concentration days in each year.

We also chose randomly 30 days in each year (150 days in total) and calculated mean ΔSLP using the same method (Fig. S2). In this case, mean ΔSLP does not show a significant difference on these randomly selected days. These results confirm that sea-level pressure 5–7 days in advance is a strong predictor for elevated PM2.5 in Chengdu. Lin et al. (2019) found that air pressure 1–2 days earlier showed a higher positive correlation with the current particle concentrations (except in winter) in the city of Lin’an in the Yangtze River Delta region. The average surface wind speed decreases with increasing sea-level pressure, which is not favorable for the horizontal dispersion of pollutants. Moreover, the high sea-level pressure may cause a weakening of the vertical diffusion of PM2.5, favoring the formation of heavy PM2.5 pollution gradually (Ning et al., 2018; Lin et al., 2019). The different lag days of our study are probably due to different distance from the sea, climate, elevation, and terrain characteristics. Further study on the relationship among cities with different topography and distance from the sea might be needed.


GAM Fitting

The GAM results were based on the training dataset, which consisted of random selection of 90% of all days. We compared the results of the two model specifications, as shown in Table 2. The only difference was that Model 1 included SLP as one of the variables whereas Model 2 included ΔSLP for 5 days in advance (SLP5d). Model 1 had the adjusted R2 = 0.70 and the variance interpretation rate was 71.7%, and Model 2 had the adjusted R2 = 0.73 and the variance interpretation rate was 74.2%. In addition, F-value of ΔSLP was 208.7 in Model 2, which is significantly higher than that in Model 1 (3.5). This indicates that ΔSLP 5 days in advance is a more important meteorological factor for PM2.5 concentrations in Chengdu compared to sea-level pressure on the same day and demonstrates that Model 2 is superior to Model 1. 

Fig. 5 shows the observed vs. model fit PM2.5 concentrations for Chengdu. It demonstrates that the GAM results are unbiased across the full range of PM2.5 concentrations. Fig. 6 shows an example of daily results from January 17, 2015, to February 10, 2015. During this period, the trends of the GAM fit values and the observed values are very similar. The square dots represent sea-level pressure of the same day. After the SLP rises to a peak, PM2.5 would also peak 5 days later. 

Fig. 5. Observed vs. model fit PM2.5 concentrations for Chengdu.Fig. 5. Observed vs. model fit PM2.5 concentrations for Chengdu.

Fig. 6. An example of daily results from January 17, 2015, to February 10, 2015.
Fig. 6. An example of daily results from January 17, 2015, to February 10, 2015.


Validation of the GAM Models

To evaluate the reliability of the GAMs we used, the R language gam.check function was used to verify the fitting results of the model. Fig. 7 shows that the points in the Q-Q plot is approximately a straight line, which indicates that the sample data and predictors were from a log-normal distribution. The frequency of residual distribution near zero is high, indicating that the fitting degree of the model is good. Fitted PM2.5 concentrations matched well with observed values. All PM2.5 values on the y-axis in Fig. 7 are on a log scale. 

Fig. 7. Residual test results of GAM: (a) normal Q-Q plot; (b) residuals vs. model-predicted values; (c) histogram of residuals; (d) fitted PM2.5 concentration values vs. observed values. All results plotted as log PM2.5 concentrations.Fig. 7. Residual test results of GAM: (a) normal Q-Q plot; (b) residuals vs. model-predicted values; (c) histogram of residuals; (d) fitted PM2.5 concentration values vs. observed values. All results plotted as log PM2.5 concentrations.

We used cross-validation (CV) to further evaluate the GAM results. The cross-validation dataset was a random selection (10%) of the entire dataset. The training data were used to build GAM Model 2, and then the remaining data were used to predict PM2.5 concentrations from the meteorological parameters. We compared R2 of training data and cross-validation data, and the residuals of predicted and observed values. We found that R2 and model residuals of the training and CV data have similar resultant values and the average of all residuals are close to zero (Table 3). 


CONCLUSIONS


We compared the daily PM2.5 concentrations at eight monitoring sites in Chengdu against a suite of meteorological parameters and discovered significant non-linear relationships. We used these meteorological parameters to develop a GAM that achieved an R2 of 0.73 in predicting the daily PM2.5 concentrations. Residual testing showed that the model is unbiased. We found that the sea-level pressure from 5 days earlier is a good predictor for PM2.5 in Chengdu, especially on days with high concentrations, enabling the government to modify emission patterns in advance and mitigate the risks of heavy pollution. Because it does not utilize emission source data, our method is easier to use and less expensive than other predictive models, such as WRF-Chem and CMAQ. However, reliable predictive variables are required for estimating the PM2.5 concentrations. Although forecasts of meteorological variables can generally be obtained on a regular basis, their uncertainty probably affects the total uncertainty. The GAM we have developed is suitable for predicting the daily PM2.5 concentration in Chengdu, a megacity in a large basin; further studies will examine its suitability for other cities with a similar topography as well as the necessary meteorological parameters for adapting it to cities with a different topography. Our novel finding that the surface pressure can predict the PM2.5 concentration 5 days later can be used to craft policies for controlling emissions in advance, thus reducing PM2.5 levels on the worst air quality days and improving public health.


ACKNOWLEDGEMENTS


This study is supported by the National Science Foundation of China (41628102, 71742004), the Program of Introducing Talents of Discipline to Universities (B08037), and PM2.5 Monitoring in the Sichuan University Campuses (SCU2015CC0001). We would like to express our gratitude to the China Meteorological Administration for providing standard hourly surface meteorological data; to the National Oceanic Atmospheric Administration, USA, for providing backward trajectory data; to the University of Wyoming, USA, for providing radiosonde meteorological data; and to the U.S. Embassy in China for providing PM2.5 monitoring data at the General Consulate in Chengdu.


REFERENCES


  1. Aldrin, M. and Haff, I.H. (2005). Generalised additive modelling of air pollution, traffic volume and meteorology. Atmos. Environ. 39: 2145–2155. [Publisher Site]

  2. Asimakopoulos, D., Flocas, H., Maggos, T. and Vasilakos, C. (2012). The role of meteorology on different sized aerosol fractions (PM10, PM2.5, PM2.5–10). Sci. Total Environ. 419: 124–135. [Publisher Site]

  3. Bertaccini, P., Dukic, V. and Ignaccolo, R. (2012). Modeling the short-term effect of traffic and meteorology on air pollution in Turin with generalized additive models. Adv. Meteorol. 2012: 609328. [Publisher Site]

  4. Bo, Z., Qiang, Z., Yang, Z., He, K., Kai, W., Zheng, G., Duan, F. Ma, Y. and Kimoto, T. (2015). Heterogeneous chemistry: a mechanism missing in current models to explain secondary inorganic aerosol formation during the January 2013 haze episode in north China. Atmos. Chem. Phys. 14: 2031–2049. [Publisher Site]

  5. Boznar, M., Lesjak, M. and Mlakar, P. (1993). A neural network-based method for short-term predictions of ambient SO2 concentrations in highly polluted industrial areas of complex terrain. Atmos. Environ. 27: 221–230. [Publisher Site]

  6. Camalier, L., Cox, W. and Dolwick, P. (2007). The effects of meteorology on ozone in urban areas and their use in assessing ozone trends. Atmos. Environ. 41: 7127–7137. [Publisher Site]

  7. Carmichael, G.R., Sandu, A., Chai, T., Daescu, D.N., Constantinescu, E.M. and Tang, Y. (2008). Predicting air quality: Improvements through advanced methods to integrate models and measurements. J. Comput. Phys. 227: 3540–3571. [Publisher Site]

  8. Chen, C., Zhu, P., Lan, L., Zhou, L., Liu, R., Sun, Q., Ban, J., Wang, W., Xu, D. and Li, T. (2018). Short-term exposures to PM2.5 and cause-specific mortality of cardiovascular health in China. Environ. Res. 161: 188–194. [Publisher Site]

  9. Cohen, A.J., Brauer, M., Burnett, R., Anderson, H.R., Frostad, J., Estep, K., Balakrishnan, K., Brunekreef, B., Dandona, L. and Dandona, R. (2017). Estimates and 25-year trends of the global burden of disease attributable to ambient air pollution: An analysis of data from the global burden of diseases study 2015. Lancet 389: 1907–1918. [Publisher Site]

  10. Di, C., Yu, S. and Bing, L. (2009). Visibility trends in six megacities in China 1973–2007. Atmos. Res. 94: 161–167. [Publisher Site]

  11. Dominici, F., McDermott, A., Zeger, S.L. and Samet, J.M. (2002). On the use of generalized additive models in time-series studies of air pollution and health. Am. J. Epidemiol. 156: 193–203.  [Publisher Site]

  12. Fang, C., Zhang, Z., Jin, M., Zou, P. and Wang, J. (2017). Pollution characteristics of PM2.5 aerosol during haze periods in Changchun, China. Aerosol Air Qual. Res. 17: 888–895. [Publisher Site]

  13. Goldberg, D.L., Gupta, P., Wang, K., Jena, C., Zhang, Y., Lu, Z. and Streets, D.G. (2019). Using gap-filled maiac AOD and WRF-Chem to estimate daily PM2.5 concentrations at 1 Km resolution in the eastern United States. Atmos. Environ. 199: 443–452. [Publisher Site]

  14. Gong, X., Kaulfus, A., Nair, U. and Jaffe, D.A. (2017). Quantifying O3 impacts in urban areas due to wildfires using a generalized additive model. Environ. Sci. Technol. 51: 13216–13223. [Publisher Site]

  15. Gong, X., Hong, S. and Jaffe, D.A. (2018). Ozone in China: spatial distribution and leading meteorological factors controlling O3 in 16 Chinese cities. Aerosol Air Qual. Res. 18: 2287–2300. [Publisher Site]

  16. Grivas, G. and Chaloulakou, A. (2006). Artificial neural network models for prediction of PM10 hourly concentrations, in the Greater Area of Athens, Greece. Atmos. Environ. 40: 1216–1229. [Publisher Site]

  17. Gustin, M.S., Fine, R., Miller, M., Dan, J. and Burley, J. (2015). The Nevada rural ozone initiative (Nvroi): Insights to understanding air pollution in complex terrain. Sci. Total Environ. 530–531: 455–470. [Publisher Site]

  18. Hastie, T.J. and Tibshirani, R.J (1990): Generalized additive models. CRC press, Florida.

  19. Health Effects Institute (2019). State of Global Air 2019. Special Report. Health Effects Institute, Boston, MA ISSN 2578-6873 © 2019 Health Effects Institute.

  20. Hoek, G., Krishnan, R.M., Beelen, R., Peters A., Ostro, B., Brunekreef, B. and Kaufman, J.D. (2013). Long-term air pollution exposure and cardio-respiratory mortality: A review. Environ. Health 12: 43. [Publisher Site]

  21. Hooyberghs, J., Mensink, C., Dumont, G., Fierens, F. and Brasseur, O. (2013). A neural network forecast for daily average PM10 concentrations in Belgium. Atmos. Environ. 39: 3279–3289. [Publisher Site]

  22. Hu, X., Ma, Z., Lin, W., Zhang, H., Hu, J., Wang, Y., Xu, X., Fuentes, J.D. and Xue, M. (2014). Impact of the Loess Plateau on the atmospheric boundary layer structure and air quality in the north China plain: a case study. Sci. Total Environ. 499: 228–237. [Publisher Site]

  23. Hübnerová, Z. and Michálek, J. (2014). Analysis of daily average PM10 predictions by generalized linear models in Brno, Czech Republic. Atmos. Pollut. Res. 5: 471–476. [Publisher Site]

  24. Jin, Y., Andersson, H. and Zhang, S. (2016). Air pollution control policies in China: A retrospective and prospects. Int. J. Environ. Res. Public Health 13: 1219. [Publisher Site]

  25. Lee, C., Wan, T., Kuo, C. and Chung, C. (2007). Modified grey model for estimating traffic tunnel air quality. Environ. Monit. Assess. 132: 351–364. [Publisher Site]

  26. Lei, R., Zhu, F., Cheng, H., Liu, J., Shen, C., Zhang, C., Xu, Y., Xiao, C., Li, X. and Zhang, J. (2019). Short-term effect of PM2.5/O3 on non-accidental and respiratory deaths in highly polluted area of China. Atmos. Pollut. Res. 10: 1412–1419. [Publisher Site]

  27. Lelieveld, J., Evans, J.S., Fnais, M., Giannadaki, D. and Pozzer, A. (2015). The contribution of outdoor air pollution sources to premature mortality on a global scale. Nature 525: 367–371. [Publisher Site]

  28. Li, T., Guo, Y., Liu, Y., Wang, J., Wang, Q., Sun, Z., He, M. and Shi, X. (2019). Estimating mortality burden attributable to short-term PM2.5 exposure: A national observational study in China. Environ. Int. 125: 245–251. [Publisher Site]

  29. Li, Y., Chen, Q., Zhao, H., Wang, L. and Tao, R. (2015). Variations in PM10, PM2.5 and PM1.0 in an urban area of the Sichuan basin and their relation to meteorological factors. Atmosphere 6: 150–163. [Publisher Site]

  30. Lin, M., Tao, J., Chan, C., Cao, J., Zhang, Z., Zhu, L. and Zhang, R. (2012). Regression analyses between recent air quality and visibility changes in megacities at four haze regions in China. Aerosol Air Qual. Res. 12: 1049–1061. [Publisher Site]

  31. Lin, X., Chen, J., Lu, T., Huang, D. and Zhang, J. (2019). Air pollution characteristics and meteorological correlates in Lin’an, Hangzhou, China. Aerosol Air Qual. Res. 19: 2770–2780. [Publisher Site]

  32. Liu, J., Han, Y., Tang, X., Zhu, J. and Zhu, T. (2016). Estimating adult mortality attributable to PM2.5 exposure in China with assimilated PM2.5 concentrations based on a ground monitoring network. Sci. Total Environ. 568: 1253–1262. [Publisher Site]

  33. Liu, M., Huang, Y., Ma, Z., Jin, Z., Liu, X., Wang, H., Liu, Y., Wang, J., Jantunen, M. and Bi, J. (2017). patial and temporal trends in the mortality burden of air pollution in China: 2004–2012. Environ. Int. 98: 75–81. [Publisher Site]

  34. Lu, W. and Wang, W. (2005). Potential assessment of the “support vector machine” method in forecasting ambient air pollutant trends. Chemosphere 59: 693–701. [Publisher Site]

  35. Luo, L., Zhang, Y., Jiang, J., Luan, H., Yu, C., Nan, P., Luo, B. and You, M. (2018). Short-term effects of ambient air pollution on hospitalization for respiratory disease in Taiyuan, China: A time-series analysis. Int. J. Environ. Res. Public Health 15: 2160. [Publisher Site]

  36. Mahowald, N. (2011). Aerosol indirect effect on biogeochemical cycles and climate. Science 334: 794–796. [Publisher Site]

  37. Miao, Y., Hu, X., Liu, S., Qian, T., Xue, M., Zheng, Y. and Wang, S. (2015). Seasonal variation of local atmospheric circulations and boundary layer structure in the Beijing-Tianjin-Hebei region and implications for air quality. J. Adv. Model. Earth Syst. 7: 1602–1626. [Publisher Site]

  38. Miao, Y., Liu, S., Guo, J., Huang, S., Yan, Y. and Lou, M. (2018). Unraveling the relationships between boundary layer height and PM2.5 pollution in China based on four-year radiosonde measurements. Environ. Pollut. 243: 1186–1195. [Publisher Site]

  39. Miao, Y., Liu, S. and Huang, S. (2019). Synoptic pattern and planetary boundary layer structure associated with aerosol pollution during winter in Beijing, China. Sci. Total Environ. 682: 464–474. [Publisher Site]

  40. Ng, K.Y. and Awang, N. (2018). Multiple linear regression and regression with time series error models in forecasting PM10 concentrations in Peninsular Malaysia. Environ. Monit. Assess. 190: 63. [Publisher Site]

  41. Ning, G., Wang, S., Yim, S.H.L., Li, J., Hu, Y., Shang, Z., Wang, J. and Wang, J. (2018). Impact of low-pressure systems on winter heavy air pollution in the northwest Sichuan Basin, China. Atmos. Chem. Phys. 18: 13601–13615. [Publisher Site]

  42. Pai, T.Y., Hanaki, K. and Chiou, R.J. (2013). Forecasting hourly roadside particulate matter in Taipei county of Taiwan based on first-order and one-variable grey model. CLEAN–Soil Air Water 41: 737–742. [Publisher Site]

  43. Palau, J., Pérez-Landa, G., Diéguez, J., Monter, C. and Millán, M. (2005). The importance of meteorological scales to forecast air pollution scenarios on coastal complex terrain. Atmos. Chem. Phys. 5: 2771–2785. [Publisher Site]

  44. Pearce, J.L., Beringer, J., Nicholls, N., Hyndman, R.J. and Tapper, N.J. (2011). Quantifying the influence of local meteorology on air quality using generalized additive models. Atmos. Environ. 45: 1328–1336. [Publisher Site]

  45. Qiao, X., Guo, H., Wang, P.F., Tang, Y., Ying, Q., Zhao, X., Deng, W.Y. and Zhang, H.L. (2019). Fine particulate matter and ozone pollution in the 18 cities of Sichuan Basin, southwestern China: Model Performance and characteristics. Aerosol Air Qual. Res. 19: 2308–2319. [Publisher Site]

  46. Ryu, Y.H., Hodzic, A., Barre, J., Descombes, G. and Minnis, P. (2018). Quantifying errors in surface ozone predictions associated with clouds over the conus: a WRF-Chem modeling study using satellite cloud retrievals. Atmos. Chem. Phys. 18: 7509–7525. [Publisher Site]

  47. Saide, P.E., Carmichael, G.R., Spak, S.N., Gallardo, L., Osses, A.E., Mena-Carrasco, M.A. and Pagowski, M. (2011). Forecasting urban PM10 and PM2.5 pollution episodes in very stable nocturnal conditions and complex terrain using WRF-Chem co tracer model. Atmos. Environ. 45: 2769–2780. [Publisher Site]

  48. Schlink, U., Herbarth, O., Richter, M., Dorling, S., Nunnari, G., Cawley, G. and Pelikan, E. (2006). Statistical models to assess the health effects and to forecast ground-level ozone. Environ. Modell. Software 21: 547–558. [Publisher Site]

  49. Sichuan Statistics Bureau (2018). Sichuan Provincial Population Statistics Bulletin 2018. [Website Link]

  50. Singh, A., Pant, P. and Pope, F.D. (2019). Air quality during and after festivals: Aerosol concentrations, composition and health effects. Atmos. Res. 227: 220–232. [Publisher Site]

  51. Sun, W., Zhang, H. and Palazoglu, A. (2013a). Prediction of 8h-average ozone concentration using a supervised hidden markov model combined with generalized linear models. Atmos. Environ. 81: 199–208. [Publisher Site]

  52. Sun, W., Zhang, H., Palazoglu, A., Singh, A., Zhang, W. and Liu, S. (2013b). Prediction of 24-hour-average PM2.5 concentrations using a hidden markov model with different emission distributions in northern California. Sci. Total Environ. 443: 93–103. [Publisher Site]

  53. Tang, G., Zhu, X., Hu, B., Xin, J., Wang, L., Münkel, C., Mao, G. and Wang, Y. (2015). Impact of emission controls on air quality in Beijing during APEC 2014: Lidar ceilometer observations. Atmos. Chem. Phys. 15: 12667–12680. [Publisher Site]

  54. Tang, G., Zhang, J., Zhu, X., Song, T., Münkel, C., Hu, B., Schäfer, K., Liu, Z., Zhang, J. and Wang, L. (2016). Mixing layer height and its implications for air pollution over Beijing, China. Atmos. Chem. Phys. 16: 2459–2475. [Publisher Site]

  55. Tao, J., Zhang, L., Engling, G., Zhang, R., Yang, Y., Cao, J., Zhu, C., Wang, Q. and Luo, L. (2013). Chemical composition of PM2.5 in an urban environment in Chengdu, China: Importance of springtime dust storms and biomass burning. Atmos. Res. 122: 270–283. [Publisher Site]

  56. Tao, J., Gao, J., Zhang, L., Zhang, R., Che, H., Zhang, Z., Lin, Z., Jing, J., Cao, J. and Hsu, S.C. (2014). PM2.5 pollution in a megacity of southwest China: Source apportionment and implication. Atmos. Chem. Phys. 14: 8679–8699. [Publisher Site]

  57. Tian, G., Qiao, Z. and Xu, X. (2014). Characteristics of particulate matter (PM10) and its relationship with meteorological factors during 2001-2012 in Beijing. Environ. Pollut. 192: 266–274. [Publisher Site]

  58. Tian, P., Cao, X., Zhang, L., Sun, N., Sun, L., Logan, T., Shi, J., Wang, Y., Ji, Y. and Lin, Y. (2017). Aerosol vertical distribution and optical properties over China from long-term satellite and ground-based remote sensing. Atmos. Chem. Phys. 17: 2509–2523. [Publisher Site]

  59. Ul-Saufie, A.Z., Yahaya, A.S., Ramli, N. and Hamid, H.A. (2012). Performance of multiple linear regression model for long-term PM10 concentration prediction based on gaseous and meteorological parameters. Applied Sci. 12: 1488–1494. [Publisher Site]

  60. Wang, M., Sampson, P.D., Hu, J., Kleeman, M., Keller, J.P., Olives, C., Szpiro, A.A., Vedal, S. and Kaufman, J.D. (2016). Combining land-use regression and chemical transport modeling in a spatio-temporal geostatistical model for ozone and PM2.5. Environ. Sci. Technol. 50: 5111–5118. [Publisher Site]

  61. Wang, P., Liu, Y., Qin, Z. and Zhang, G. (2014). A novel hybrid forecasting model for PM10 and SO2 daily concentrations. Sci. Total Environ. 505: 1202–1212. [Publisher Site]

  62. Wilkins, J.L., Pouliot, G., Foley, K., Appel, W. and Pierce, T. (2018). The impact of US wildland fires on ozone and particulate matter: A comparison of measurements and CMAQ model predictions from 2008 to 2012. Int. J. Wildland Fire 27: 684–698. [Publisher Site]

  63. Wood, S.N. (2006). Generalized Additive Models: An Introduction with R. Chapman & Hall/CRC, Boca Raton, Florida. [Publisher Site]

  64. Wu, H., Tang, X., Wang, Z., Wu, L., Lu, M., Wei, L. and Zhu, J. (2018). Probabilistic automatic outlier detection for surface air quality measurements from the China national environmental monitoring network. Adv. Atmos. Sci. 35: 1522–1532. [Publisher Site]

  65. Xue, T., Zheng, Y., Tong, D., Zheng, B., Li, X., Zhu, T. and Zhang, Q. (2019). Spatiotemporal continuous estimates of PM2.5 concentrations in China, 2000–2016: A machine learning method with inputs from satellites, chemical transport model, and ground observations. Environ. Int. 123: 345–357. [Publisher Site]

  66. Yang, X., Wu, Q., Zhao, R., Cheng, H., He, H., Ma, Q., Wang, L. and Luo, H. (2019). New method for evaluating winter air quality: PM2.5 assessment using community multi-scale air quality modeling (CMAQ) in Xi'an. Atmos. Environ. 211: 18–28. [Publisher Site]

  67. Zeng, S. and Zhang, Y. (2017). The effect of meteorological elements on continuing heavy air pollution: A case study in the Chengdu area during the 2014 spring festival. Atmosphere 8:71. [Publisher Site]

  68. Zeng, Y., Cao, Y., Qiao, X., Seyler, B.C. and Tang, Y. (2019). Air pollution reduction in China: Recent success but great challenge for the future. Sci. Total Environ. 663: 329–337. [Publisher Site]

  69. Zhang, H., Zhang, W., Palazoglu, A. and Sun, W. (2012). Prediction of ozone levels using a hidden Markov model (HMM) with Gamma distribution. Atmos. Environ. 62: 64–73. [Publisher Site]

  70. Zhao, D., Chen, H., Sun, X. and Shi, Z. (2018). Spatio-temporal variation of PM2.5 pollution and its relationship with meteorology among five megacities in China. Aerosol Air Qual. Res. 18: 2318–2331. [Publisher Site]


Share this article with your colleagues 

 

Subscribe to our Newsletter 

Aerosol and Air Quality Research has published over 2,000 peer-reviewed articles. Enter your email address to receive latest updates and research articles to your inbox every second week.

7.3
2022CiteScore
 
 
77st percentile
Powered by
Scopus
 
   SCImago Journal & Country Rank

2022 Impact Factor: 4.0
5-Year Impact Factor: 3.4

Call for Papers for the special issue on: "Carbonaceous Aerosols in the Atmosphere"

Aerosol and Air Quality Research partners with Publons

CLOCKSS system has permission to ingest, preserve, and serve this Archival Unit
CLOCKSS system has permission to ingest, preserve, and serve this Archival Unit

Aerosol and Air Quality Research (AAQR) is an independently-run non-profit journal that promotes submissions of high-quality research and strives to be one of the leading aerosol and air quality open-access journals in the world. We use cookies on this website to personalize content to improve your user experience and analyze our traffic. By using this site you agree to its use of cookies.